]>
Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * fs/dcache.c | |
4 | * | |
5 | * Complete reimplementation | |
6 | * (C) 1997 Thomas Schoebel-Theuer, | |
7 | * with heavy changes by Linus Torvalds | |
8 | */ | |
9 | ||
10 | /* | |
11 | * Notes on the allocation strategy: | |
12 | * | |
13 | * The dcache is a master of the icache - whenever a dcache entry | |
14 | * exists, the inode will always exist. "iput()" is done either when | |
15 | * the dcache entry is deleted or garbage collected. | |
16 | */ | |
17 | ||
7a5cf791 | 18 | #include <linux/ratelimit.h> |
1da177e4 LT |
19 | #include <linux/string.h> |
20 | #include <linux/mm.h> | |
21 | #include <linux/fs.h> | |
0bf3d5c1 | 22 | #include <linux/fscrypt.h> |
7a91bf7f | 23 | #include <linux/fsnotify.h> |
1da177e4 LT |
24 | #include <linux/slab.h> |
25 | #include <linux/init.h> | |
1da177e4 LT |
26 | #include <linux/hash.h> |
27 | #include <linux/cache.h> | |
630d9c47 | 28 | #include <linux/export.h> |
1da177e4 LT |
29 | #include <linux/security.h> |
30 | #include <linux/seqlock.h> | |
57c8a661 | 31 | #include <linux/memblock.h> |
ceb5bdc2 NP |
32 | #include <linux/bit_spinlock.h> |
33 | #include <linux/rculist_bl.h> | |
f6041567 | 34 | #include <linux/list_lru.h> |
07f3f05c | 35 | #include "internal.h" |
b2dba1af | 36 | #include "mount.h" |
1da177e4 | 37 | |
789680d1 NP |
38 | /* |
39 | * Usage: | |
873feea0 | 40 | * dcache->d_inode->i_lock protects: |
946e51f2 | 41 | * - i_dentry, d_u.d_alias, d_inode of aliases |
ceb5bdc2 NP |
42 | * dcache_hash_bucket lock protects: |
43 | * - the dcache hash table | |
f1ee6162 N |
44 | * s_roots bl list spinlock protects: |
45 | * - the s_roots list (see __d_drop) | |
19156840 | 46 | * dentry->d_sb->s_dentry_lru_lock protects: |
23044507 NP |
47 | * - the dcache lru lists and counters |
48 | * d_lock protects: | |
49 | * - d_flags | |
50 | * - d_name | |
51 | * - d_lru | |
b7ab39f6 | 52 | * - d_count |
da502956 | 53 | * - d_unhashed() |
2fd6b7f5 NP |
54 | * - d_parent and d_subdirs |
55 | * - childrens' d_child and d_parent | |
946e51f2 | 56 | * - d_u.d_alias, d_inode |
789680d1 NP |
57 | * |
58 | * Ordering: | |
873feea0 | 59 | * dentry->d_inode->i_lock |
b5c84bf6 | 60 | * dentry->d_lock |
19156840 | 61 | * dentry->d_sb->s_dentry_lru_lock |
ceb5bdc2 | 62 | * dcache_hash_bucket lock |
f1ee6162 | 63 | * s_roots lock |
789680d1 | 64 | * |
da502956 NP |
65 | * If there is an ancestor relationship: |
66 | * dentry->d_parent->...->d_parent->d_lock | |
67 | * ... | |
68 | * dentry->d_parent->d_lock | |
69 | * dentry->d_lock | |
70 | * | |
71 | * If no ancestor relationship: | |
076515fc | 72 | * arbitrary, since it's serialized on rename_lock |
789680d1 | 73 | */ |
fa3536cc | 74 | int sysctl_vfs_cache_pressure __read_mostly = 100; |
1da177e4 LT |
75 | EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure); |
76 | ||
74c3cbe3 | 77 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock); |
1da177e4 | 78 | |
949854d0 | 79 | EXPORT_SYMBOL(rename_lock); |
1da177e4 | 80 | |
68279f9c | 81 | static struct kmem_cache *dentry_cache __ro_after_init; |
1da177e4 | 82 | |
cdf01226 DH |
83 | const struct qstr empty_name = QSTR_INIT("", 0); |
84 | EXPORT_SYMBOL(empty_name); | |
85 | const struct qstr slash_name = QSTR_INIT("/", 1); | |
86 | EXPORT_SYMBOL(slash_name); | |
80e5d1ff AV |
87 | const struct qstr dotdot_name = QSTR_INIT("..", 2); |
88 | EXPORT_SYMBOL(dotdot_name); | |
cdf01226 | 89 | |
1da177e4 LT |
90 | /* |
91 | * This is the single most critical data structure when it comes | |
92 | * to the dcache: the hashtable for lookups. Somebody should try | |
93 | * to make this good - I've just made it work. | |
94 | * | |
95 | * This hash-function tries to avoid losing too many bits of hash | |
96 | * information, yet avoid using a prime hash-size or similar. | |
97 | */ | |
1da177e4 | 98 | |
68279f9c | 99 | static unsigned int d_hash_shift __ro_after_init; |
ceb5bdc2 | 100 | |
68279f9c | 101 | static struct hlist_bl_head *dentry_hashtable __ro_after_init; |
ceb5bdc2 | 102 | |
8387ff25 | 103 | static inline struct hlist_bl_head *d_hash(unsigned int hash) |
ceb5bdc2 | 104 | { |
854d3e63 | 105 | return dentry_hashtable + (hash >> d_hash_shift); |
ceb5bdc2 NP |
106 | } |
107 | ||
94bdd655 AV |
108 | #define IN_LOOKUP_SHIFT 10 |
109 | static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT]; | |
110 | ||
111 | static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent, | |
112 | unsigned int hash) | |
113 | { | |
114 | hash += (unsigned long) parent / L1_CACHE_BYTES; | |
115 | return in_lookup_hashtable + hash_32(hash, IN_LOOKUP_SHIFT); | |
116 | } | |
117 | ||
c8c0c239 LC |
118 | struct dentry_stat_t { |
119 | long nr_dentry; | |
120 | long nr_unused; | |
121 | long age_limit; /* age in seconds */ | |
122 | long want_pages; /* pages requested by system */ | |
123 | long nr_negative; /* # of unused negative dentries */ | |
124 | long dummy; /* Reserved for future use */ | |
1da177e4 LT |
125 | }; |
126 | ||
3942c07c | 127 | static DEFINE_PER_CPU(long, nr_dentry); |
62d36c77 | 128 | static DEFINE_PER_CPU(long, nr_dentry_unused); |
af0c9af1 | 129 | static DEFINE_PER_CPU(long, nr_dentry_negative); |
312d3ca8 CH |
130 | |
131 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) | |
c8c0c239 LC |
132 | /* Statistics gathering. */ |
133 | static struct dentry_stat_t dentry_stat = { | |
134 | .age_limit = 45, | |
135 | }; | |
62d36c77 DC |
136 | |
137 | /* | |
138 | * Here we resort to our own counters instead of using generic per-cpu counters | |
139 | * for consistency with what the vfs inode code does. We are expected to harvest | |
140 | * better code and performance by having our own specialized counters. | |
141 | * | |
142 | * Please note that the loop is done over all possible CPUs, not over all online | |
143 | * CPUs. The reason for this is that we don't want to play games with CPUs going | |
144 | * on and off. If one of them goes off, we will just keep their counters. | |
145 | * | |
146 | * glommer: See cffbc8a for details, and if you ever intend to change this, | |
147 | * please update all vfs counters to match. | |
148 | */ | |
3942c07c | 149 | static long get_nr_dentry(void) |
3e880fb5 NP |
150 | { |
151 | int i; | |
3942c07c | 152 | long sum = 0; |
3e880fb5 NP |
153 | for_each_possible_cpu(i) |
154 | sum += per_cpu(nr_dentry, i); | |
155 | return sum < 0 ? 0 : sum; | |
156 | } | |
157 | ||
62d36c77 DC |
158 | static long get_nr_dentry_unused(void) |
159 | { | |
160 | int i; | |
161 | long sum = 0; | |
162 | for_each_possible_cpu(i) | |
163 | sum += per_cpu(nr_dentry_unused, i); | |
164 | return sum < 0 ? 0 : sum; | |
165 | } | |
166 | ||
af0c9af1 WL |
167 | static long get_nr_dentry_negative(void) |
168 | { | |
169 | int i; | |
170 | long sum = 0; | |
171 | ||
172 | for_each_possible_cpu(i) | |
173 | sum += per_cpu(nr_dentry_negative, i); | |
174 | return sum < 0 ? 0 : sum; | |
175 | } | |
176 | ||
c8c0c239 LC |
177 | static int proc_nr_dentry(struct ctl_table *table, int write, void *buffer, |
178 | size_t *lenp, loff_t *ppos) | |
312d3ca8 | 179 | { |
3e880fb5 | 180 | dentry_stat.nr_dentry = get_nr_dentry(); |
62d36c77 | 181 | dentry_stat.nr_unused = get_nr_dentry_unused(); |
af0c9af1 | 182 | dentry_stat.nr_negative = get_nr_dentry_negative(); |
3942c07c | 183 | return proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
312d3ca8 | 184 | } |
c8c0c239 LC |
185 | |
186 | static struct ctl_table fs_dcache_sysctls[] = { | |
187 | { | |
188 | .procname = "dentry-state", | |
189 | .data = &dentry_stat, | |
190 | .maxlen = 6*sizeof(long), | |
191 | .mode = 0444, | |
192 | .proc_handler = proc_nr_dentry, | |
193 | }, | |
194 | { } | |
195 | }; | |
196 | ||
197 | static int __init init_fs_dcache_sysctls(void) | |
198 | { | |
199 | register_sysctl_init("fs", fs_dcache_sysctls); | |
200 | return 0; | |
201 | } | |
202 | fs_initcall(init_fs_dcache_sysctls); | |
312d3ca8 CH |
203 | #endif |
204 | ||
5483f18e LT |
205 | /* |
206 | * Compare 2 name strings, return 0 if they match, otherwise non-zero. | |
207 | * The strings are both count bytes long, and count is non-zero. | |
208 | */ | |
e419b4cc LT |
209 | #ifdef CONFIG_DCACHE_WORD_ACCESS |
210 | ||
211 | #include <asm/word-at-a-time.h> | |
212 | /* | |
213 | * NOTE! 'cs' and 'scount' come from a dentry, so it has a | |
214 | * aligned allocation for this particular component. We don't | |
215 | * strictly need the load_unaligned_zeropad() safety, but it | |
216 | * doesn't hurt either. | |
217 | * | |
218 | * In contrast, 'ct' and 'tcount' can be from a pathname, and do | |
219 | * need the careful unaligned handling. | |
220 | */ | |
94753db5 | 221 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
5483f18e | 222 | { |
bfcfaa77 | 223 | unsigned long a,b,mask; |
bfcfaa77 LT |
224 | |
225 | for (;;) { | |
bfe7aa6c | 226 | a = read_word_at_a_time(cs); |
e419b4cc | 227 | b = load_unaligned_zeropad(ct); |
bfcfaa77 LT |
228 | if (tcount < sizeof(unsigned long)) |
229 | break; | |
230 | if (unlikely(a != b)) | |
231 | return 1; | |
232 | cs += sizeof(unsigned long); | |
233 | ct += sizeof(unsigned long); | |
234 | tcount -= sizeof(unsigned long); | |
235 | if (!tcount) | |
236 | return 0; | |
237 | } | |
a5c21dce | 238 | mask = bytemask_from_count(tcount); |
bfcfaa77 | 239 | return unlikely(!!((a ^ b) & mask)); |
e419b4cc LT |
240 | } |
241 | ||
bfcfaa77 | 242 | #else |
e419b4cc | 243 | |
94753db5 | 244 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
e419b4cc | 245 | { |
5483f18e LT |
246 | do { |
247 | if (*cs != *ct) | |
248 | return 1; | |
249 | cs++; | |
250 | ct++; | |
251 | tcount--; | |
252 | } while (tcount); | |
253 | return 0; | |
254 | } | |
255 | ||
e419b4cc LT |
256 | #endif |
257 | ||
94753db5 LT |
258 | static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount) |
259 | { | |
94753db5 LT |
260 | /* |
261 | * Be careful about RCU walk racing with rename: | |
506458ef | 262 | * use 'READ_ONCE' to fetch the name pointer. |
94753db5 LT |
263 | * |
264 | * NOTE! Even if a rename will mean that the length | |
265 | * was not loaded atomically, we don't care. The | |
266 | * RCU walk will check the sequence count eventually, | |
267 | * and catch it. And we won't overrun the buffer, | |
268 | * because we're reading the name pointer atomically, | |
269 | * and a dentry name is guaranteed to be properly | |
270 | * terminated with a NUL byte. | |
271 | * | |
272 | * End result: even if 'len' is wrong, we'll exit | |
273 | * early because the data cannot match (there can | |
274 | * be no NUL in the ct/tcount data) | |
275 | */ | |
506458ef | 276 | const unsigned char *cs = READ_ONCE(dentry->d_name.name); |
ae0a843c | 277 | |
6326c71f | 278 | return dentry_string_cmp(cs, ct, tcount); |
94753db5 LT |
279 | } |
280 | ||
8d85b484 AV |
281 | struct external_name { |
282 | union { | |
283 | atomic_t count; | |
284 | struct rcu_head head; | |
285 | } u; | |
286 | unsigned char name[]; | |
287 | }; | |
288 | ||
289 | static inline struct external_name *external_name(struct dentry *dentry) | |
290 | { | |
291 | return container_of(dentry->d_name.name, struct external_name, name[0]); | |
292 | } | |
293 | ||
9c82ab9c | 294 | static void __d_free(struct rcu_head *head) |
1da177e4 | 295 | { |
9c82ab9c CH |
296 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); |
297 | ||
8d85b484 AV |
298 | kmem_cache_free(dentry_cache, dentry); |
299 | } | |
300 | ||
301 | static void __d_free_external(struct rcu_head *head) | |
302 | { | |
303 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); | |
2e03b4bc | 304 | kfree(external_name(dentry)); |
f1782c9b | 305 | kmem_cache_free(dentry_cache, dentry); |
1da177e4 LT |
306 | } |
307 | ||
810bb172 AV |
308 | static inline int dname_external(const struct dentry *dentry) |
309 | { | |
310 | return dentry->d_name.name != dentry->d_iname; | |
311 | } | |
312 | ||
49d31c2f AV |
313 | void take_dentry_name_snapshot(struct name_snapshot *name, struct dentry *dentry) |
314 | { | |
315 | spin_lock(&dentry->d_lock); | |
230c6402 | 316 | name->name = dentry->d_name; |
49d31c2f | 317 | if (unlikely(dname_external(dentry))) { |
230c6402 | 318 | atomic_inc(&external_name(dentry)->u.count); |
49d31c2f | 319 | } else { |
6cd00a01 TH |
320 | memcpy(name->inline_name, dentry->d_iname, |
321 | dentry->d_name.len + 1); | |
230c6402 | 322 | name->name.name = name->inline_name; |
49d31c2f | 323 | } |
230c6402 | 324 | spin_unlock(&dentry->d_lock); |
49d31c2f AV |
325 | } |
326 | EXPORT_SYMBOL(take_dentry_name_snapshot); | |
327 | ||
328 | void release_dentry_name_snapshot(struct name_snapshot *name) | |
329 | { | |
230c6402 | 330 | if (unlikely(name->name.name != name->inline_name)) { |
49d31c2f | 331 | struct external_name *p; |
230c6402 | 332 | p = container_of(name->name.name, struct external_name, name[0]); |
49d31c2f | 333 | if (unlikely(atomic_dec_and_test(&p->u.count))) |
2e03b4bc | 334 | kfree_rcu(p, u.head); |
49d31c2f AV |
335 | } |
336 | } | |
337 | EXPORT_SYMBOL(release_dentry_name_snapshot); | |
338 | ||
4bf46a27 DH |
339 | static inline void __d_set_inode_and_type(struct dentry *dentry, |
340 | struct inode *inode, | |
341 | unsigned type_flags) | |
342 | { | |
343 | unsigned flags; | |
344 | ||
345 | dentry->d_inode = inode; | |
4bf46a27 DH |
346 | flags = READ_ONCE(dentry->d_flags); |
347 | flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU); | |
348 | flags |= type_flags; | |
2fa6b1e0 | 349 | smp_store_release(&dentry->d_flags, flags); |
4bf46a27 DH |
350 | } |
351 | ||
4bf46a27 DH |
352 | static inline void __d_clear_type_and_inode(struct dentry *dentry) |
353 | { | |
354 | unsigned flags = READ_ONCE(dentry->d_flags); | |
355 | ||
356 | flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU); | |
357 | WRITE_ONCE(dentry->d_flags, flags); | |
4bf46a27 | 358 | dentry->d_inode = NULL; |
af0c9af1 WL |
359 | if (dentry->d_flags & DCACHE_LRU_LIST) |
360 | this_cpu_inc(nr_dentry_negative); | |
4bf46a27 DH |
361 | } |
362 | ||
b4f0354e AV |
363 | static void dentry_free(struct dentry *dentry) |
364 | { | |
946e51f2 | 365 | WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias)); |
8d85b484 AV |
366 | if (unlikely(dname_external(dentry))) { |
367 | struct external_name *p = external_name(dentry); | |
368 | if (likely(atomic_dec_and_test(&p->u.count))) { | |
369 | call_rcu(&dentry->d_u.d_rcu, __d_free_external); | |
370 | return; | |
371 | } | |
372 | } | |
b4f0354e | 373 | /* if dentry was never visible to RCU, immediate free is OK */ |
5467a68c | 374 | if (dentry->d_flags & DCACHE_NORCU) |
b4f0354e AV |
375 | __d_free(&dentry->d_u.d_rcu); |
376 | else | |
377 | call_rcu(&dentry->d_u.d_rcu, __d_free); | |
378 | } | |
379 | ||
1da177e4 LT |
380 | /* |
381 | * Release the dentry's inode, using the filesystem | |
550dce01 | 382 | * d_iput() operation if defined. |
31e6b01f NP |
383 | */ |
384 | static void dentry_unlink_inode(struct dentry * dentry) | |
385 | __releases(dentry->d_lock) | |
873feea0 | 386 | __releases(dentry->d_inode->i_lock) |
31e6b01f NP |
387 | { |
388 | struct inode *inode = dentry->d_inode; | |
a528aca7 | 389 | |
4c0d7cd5 | 390 | raw_write_seqcount_begin(&dentry->d_seq); |
4bf46a27 | 391 | __d_clear_type_and_inode(dentry); |
946e51f2 | 392 | hlist_del_init(&dentry->d_u.d_alias); |
4c0d7cd5 | 393 | raw_write_seqcount_end(&dentry->d_seq); |
31e6b01f | 394 | spin_unlock(&dentry->d_lock); |
873feea0 | 395 | spin_unlock(&inode->i_lock); |
31e6b01f NP |
396 | if (!inode->i_nlink) |
397 | fsnotify_inoderemove(inode); | |
398 | if (dentry->d_op && dentry->d_op->d_iput) | |
399 | dentry->d_op->d_iput(dentry, inode); | |
400 | else | |
401 | iput(inode); | |
402 | } | |
403 | ||
89dc77bc LT |
404 | /* |
405 | * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry | |
406 | * is in use - which includes both the "real" per-superblock | |
407 | * LRU list _and_ the DCACHE_SHRINK_LIST use. | |
408 | * | |
409 | * The DCACHE_SHRINK_LIST bit is set whenever the dentry is | |
410 | * on the shrink list (ie not on the superblock LRU list). | |
411 | * | |
412 | * The per-cpu "nr_dentry_unused" counters are updated with | |
413 | * the DCACHE_LRU_LIST bit. | |
414 | * | |
af0c9af1 WL |
415 | * The per-cpu "nr_dentry_negative" counters are only updated |
416 | * when deleted from or added to the per-superblock LRU list, not | |
417 | * from/to the shrink list. That is to avoid an unneeded dec/inc | |
418 | * pair when moving from LRU to shrink list in select_collect(). | |
419 | * | |
89dc77bc LT |
420 | * These helper functions make sure we always follow the |
421 | * rules. d_lock must be held by the caller. | |
422 | */ | |
423 | #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x)) | |
424 | static void d_lru_add(struct dentry *dentry) | |
425 | { | |
426 | D_FLAG_VERIFY(dentry, 0); | |
427 | dentry->d_flags |= DCACHE_LRU_LIST; | |
428 | this_cpu_inc(nr_dentry_unused); | |
af0c9af1 WL |
429 | if (d_is_negative(dentry)) |
430 | this_cpu_inc(nr_dentry_negative); | |
0a97c01c NP |
431 | WARN_ON_ONCE(!list_lru_add_obj( |
432 | &dentry->d_sb->s_dentry_lru, &dentry->d_lru)); | |
89dc77bc LT |
433 | } |
434 | ||
435 | static void d_lru_del(struct dentry *dentry) | |
436 | { | |
437 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
438 | dentry->d_flags &= ~DCACHE_LRU_LIST; | |
439 | this_cpu_dec(nr_dentry_unused); | |
af0c9af1 WL |
440 | if (d_is_negative(dentry)) |
441 | this_cpu_dec(nr_dentry_negative); | |
0a97c01c NP |
442 | WARN_ON_ONCE(!list_lru_del_obj( |
443 | &dentry->d_sb->s_dentry_lru, &dentry->d_lru)); | |
89dc77bc LT |
444 | } |
445 | ||
446 | static void d_shrink_del(struct dentry *dentry) | |
447 | { | |
448 | D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); | |
449 | list_del_init(&dentry->d_lru); | |
450 | dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST); | |
451 | this_cpu_dec(nr_dentry_unused); | |
452 | } | |
453 | ||
454 | static void d_shrink_add(struct dentry *dentry, struct list_head *list) | |
455 | { | |
456 | D_FLAG_VERIFY(dentry, 0); | |
457 | list_add(&dentry->d_lru, list); | |
458 | dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST; | |
459 | this_cpu_inc(nr_dentry_unused); | |
460 | } | |
461 | ||
462 | /* | |
463 | * These can only be called under the global LRU lock, ie during the | |
464 | * callback for freeing the LRU list. "isolate" removes it from the | |
465 | * LRU lists entirely, while shrink_move moves it to the indicated | |
466 | * private list. | |
467 | */ | |
3f97b163 | 468 | static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry) |
89dc77bc LT |
469 | { |
470 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
471 | dentry->d_flags &= ~DCACHE_LRU_LIST; | |
472 | this_cpu_dec(nr_dentry_unused); | |
af0c9af1 WL |
473 | if (d_is_negative(dentry)) |
474 | this_cpu_dec(nr_dentry_negative); | |
3f97b163 | 475 | list_lru_isolate(lru, &dentry->d_lru); |
89dc77bc LT |
476 | } |
477 | ||
3f97b163 VD |
478 | static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry, |
479 | struct list_head *list) | |
89dc77bc LT |
480 | { |
481 | D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST); | |
482 | dentry->d_flags |= DCACHE_SHRINK_LIST; | |
af0c9af1 WL |
483 | if (d_is_negative(dentry)) |
484 | this_cpu_dec(nr_dentry_negative); | |
3f97b163 | 485 | list_lru_isolate_move(lru, &dentry->d_lru, list); |
89dc77bc LT |
486 | } |
487 | ||
61647823 | 488 | static void ___d_drop(struct dentry *dentry) |
789680d1 | 489 | { |
0632a9ac AV |
490 | struct hlist_bl_head *b; |
491 | /* | |
492 | * Hashed dentries are normally on the dentry hashtable, | |
493 | * with the exception of those newly allocated by | |
494 | * d_obtain_root, which are always IS_ROOT: | |
495 | */ | |
496 | if (unlikely(IS_ROOT(dentry))) | |
497 | b = &dentry->d_sb->s_roots; | |
498 | else | |
499 | b = d_hash(dentry->d_name.hash); | |
b61625d2 | 500 | |
0632a9ac AV |
501 | hlist_bl_lock(b); |
502 | __hlist_bl_del(&dentry->d_hash); | |
503 | hlist_bl_unlock(b); | |
789680d1 | 504 | } |
61647823 N |
505 | |
506 | void __d_drop(struct dentry *dentry) | |
507 | { | |
0632a9ac AV |
508 | if (!d_unhashed(dentry)) { |
509 | ___d_drop(dentry); | |
510 | dentry->d_hash.pprev = NULL; | |
511 | write_seqcount_invalidate(&dentry->d_seq); | |
512 | } | |
61647823 | 513 | } |
789680d1 NP |
514 | EXPORT_SYMBOL(__d_drop); |
515 | ||
961f3c89 MCC |
516 | /** |
517 | * d_drop - drop a dentry | |
518 | * @dentry: dentry to drop | |
519 | * | |
520 | * d_drop() unhashes the entry from the parent dentry hashes, so that it won't | |
521 | * be found through a VFS lookup any more. Note that this is different from | |
522 | * deleting the dentry - d_delete will try to mark the dentry negative if | |
523 | * possible, giving a successful _negative_ lookup, while d_drop will | |
524 | * just make the cache lookup fail. | |
525 | * | |
526 | * d_drop() is used mainly for stuff that wants to invalidate a dentry for some | |
527 | * reason (NFS timeouts or autofs deletes). | |
528 | * | |
529 | * __d_drop requires dentry->d_lock | |
530 | * | |
531 | * ___d_drop doesn't mark dentry as "unhashed" | |
532 | * (dentry->d_hash.pprev will be LIST_POISON2, not NULL). | |
533 | */ | |
789680d1 NP |
534 | void d_drop(struct dentry *dentry) |
535 | { | |
789680d1 NP |
536 | spin_lock(&dentry->d_lock); |
537 | __d_drop(dentry); | |
538 | spin_unlock(&dentry->d_lock); | |
789680d1 NP |
539 | } |
540 | EXPORT_SYMBOL(d_drop); | |
541 | ||
ba65dc5e AV |
542 | static inline void dentry_unlist(struct dentry *dentry, struct dentry *parent) |
543 | { | |
544 | struct dentry *next; | |
545 | /* | |
546 | * Inform d_walk() and shrink_dentry_list() that we are no longer | |
547 | * attached to the dentry tree | |
548 | */ | |
549 | dentry->d_flags |= DCACHE_DENTRY_KILLED; | |
550 | if (unlikely(list_empty(&dentry->d_child))) | |
551 | return; | |
552 | __list_del_entry(&dentry->d_child); | |
553 | /* | |
554 | * Cursors can move around the list of children. While we'd been | |
555 | * a normal list member, it didn't matter - ->d_child.next would've | |
556 | * been updated. However, from now on it won't be and for the | |
557 | * things like d_walk() it might end up with a nasty surprise. | |
558 | * Normally d_walk() doesn't care about cursors moving around - | |
559 | * ->d_lock on parent prevents that and since a cursor has no children | |
560 | * of its own, we get through it without ever unlocking the parent. | |
561 | * There is one exception, though - if we ascend from a child that | |
562 | * gets killed as soon as we unlock it, the next sibling is found | |
563 | * using the value left in its ->d_child.next. And if _that_ | |
564 | * pointed to a cursor, and cursor got moved (e.g. by lseek()) | |
565 | * before d_walk() regains parent->d_lock, we'll end up skipping | |
566 | * everything the cursor had been moved past. | |
567 | * | |
568 | * Solution: make sure that the pointer left behind in ->d_child.next | |
569 | * points to something that won't be moving around. I.e. skip the | |
570 | * cursors. | |
571 | */ | |
572 | while (dentry->d_child.next != &parent->d_subdirs) { | |
573 | next = list_entry(dentry->d_child.next, struct dentry, d_child); | |
574 | if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR))) | |
575 | break; | |
576 | dentry->d_child.next = next->d_child.next; | |
577 | } | |
578 | } | |
579 | ||
e55fd011 | 580 | static void __dentry_kill(struct dentry *dentry) |
77812a1e | 581 | { |
41edf278 AV |
582 | struct dentry *parent = NULL; |
583 | bool can_free = true; | |
41edf278 | 584 | if (!IS_ROOT(dentry)) |
77812a1e | 585 | parent = dentry->d_parent; |
31e6b01f | 586 | |
0d98439e LT |
587 | /* |
588 | * The dentry is now unrecoverably dead to the world. | |
589 | */ | |
590 | lockref_mark_dead(&dentry->d_lockref); | |
591 | ||
f0023bc6 | 592 | /* |
f0023bc6 SW |
593 | * inform the fs via d_prune that this dentry is about to be |
594 | * unhashed and destroyed. | |
595 | */ | |
29266201 | 596 | if (dentry->d_flags & DCACHE_OP_PRUNE) |
61572bb1 YZ |
597 | dentry->d_op->d_prune(dentry); |
598 | ||
01b60351 AV |
599 | if (dentry->d_flags & DCACHE_LRU_LIST) { |
600 | if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) | |
601 | d_lru_del(dentry); | |
01b60351 | 602 | } |
77812a1e NP |
603 | /* if it was on the hash then remove it */ |
604 | __d_drop(dentry); | |
ba65dc5e | 605 | dentry_unlist(dentry, parent); |
03b3b889 AV |
606 | if (parent) |
607 | spin_unlock(&parent->d_lock); | |
550dce01 AV |
608 | if (dentry->d_inode) |
609 | dentry_unlink_inode(dentry); | |
610 | else | |
611 | spin_unlock(&dentry->d_lock); | |
03b3b889 AV |
612 | this_cpu_dec(nr_dentry); |
613 | if (dentry->d_op && dentry->d_op->d_release) | |
614 | dentry->d_op->d_release(dentry); | |
615 | ||
41edf278 AV |
616 | spin_lock(&dentry->d_lock); |
617 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { | |
618 | dentry->d_flags |= DCACHE_MAY_FREE; | |
619 | can_free = false; | |
620 | } | |
621 | spin_unlock(&dentry->d_lock); | |
41edf278 AV |
622 | if (likely(can_free)) |
623 | dentry_free(dentry); | |
9c5f1d30 | 624 | cond_resched(); |
e55fd011 AV |
625 | } |
626 | ||
8b987a46 | 627 | static struct dentry *__lock_parent(struct dentry *dentry) |
046b961b | 628 | { |
8b987a46 | 629 | struct dentry *parent; |
046b961b | 630 | rcu_read_lock(); |
c2338f2d | 631 | spin_unlock(&dentry->d_lock); |
046b961b | 632 | again: |
66702eb5 | 633 | parent = READ_ONCE(dentry->d_parent); |
046b961b AV |
634 | spin_lock(&parent->d_lock); |
635 | /* | |
636 | * We can't blindly lock dentry until we are sure | |
637 | * that we won't violate the locking order. | |
638 | * Any changes of dentry->d_parent must have | |
639 | * been done with parent->d_lock held, so | |
640 | * spin_lock() above is enough of a barrier | |
641 | * for checking if it's still our child. | |
642 | */ | |
643 | if (unlikely(parent != dentry->d_parent)) { | |
644 | spin_unlock(&parent->d_lock); | |
645 | goto again; | |
646 | } | |
65d8eb5a AV |
647 | rcu_read_unlock(); |
648 | if (parent != dentry) | |
9f12600f | 649 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
65d8eb5a | 650 | else |
046b961b AV |
651 | parent = NULL; |
652 | return parent; | |
653 | } | |
654 | ||
8b987a46 AV |
655 | static inline struct dentry *lock_parent(struct dentry *dentry) |
656 | { | |
657 | struct dentry *parent = dentry->d_parent; | |
658 | if (IS_ROOT(dentry)) | |
659 | return NULL; | |
660 | if (likely(spin_trylock(&parent->d_lock))) | |
661 | return parent; | |
662 | return __lock_parent(dentry); | |
663 | } | |
664 | ||
a338579f AV |
665 | static inline bool retain_dentry(struct dentry *dentry) |
666 | { | |
667 | WARN_ON(d_in_lookup(dentry)); | |
668 | ||
669 | /* Unreachable? Get rid of it */ | |
670 | if (unlikely(d_unhashed(dentry))) | |
671 | return false; | |
672 | ||
673 | if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED)) | |
674 | return false; | |
675 | ||
676 | if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) { | |
677 | if (dentry->d_op->d_delete(dentry)) | |
678 | return false; | |
679 | } | |
2c567af4 IW |
680 | |
681 | if (unlikely(dentry->d_flags & DCACHE_DONTCACHE)) | |
682 | return false; | |
683 | ||
62d9956c AV |
684 | /* retain; LRU fodder */ |
685 | dentry->d_lockref.count--; | |
686 | if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST))) | |
687 | d_lru_add(dentry); | |
688 | else if (unlikely(!(dentry->d_flags & DCACHE_REFERENCED))) | |
689 | dentry->d_flags |= DCACHE_REFERENCED; | |
a338579f AV |
690 | return true; |
691 | } | |
692 | ||
2c567af4 IW |
693 | void d_mark_dontcache(struct inode *inode) |
694 | { | |
695 | struct dentry *de; | |
696 | ||
697 | spin_lock(&inode->i_lock); | |
698 | hlist_for_each_entry(de, &inode->i_dentry, d_u.d_alias) { | |
699 | spin_lock(&de->d_lock); | |
700 | de->d_flags |= DCACHE_DONTCACHE; | |
701 | spin_unlock(&de->d_lock); | |
702 | } | |
703 | inode->i_state |= I_DONTCACHE; | |
704 | spin_unlock(&inode->i_lock); | |
705 | } | |
706 | EXPORT_SYMBOL(d_mark_dontcache); | |
707 | ||
c1d0c1a2 JO |
708 | /* |
709 | * Finish off a dentry we've decided to kill. | |
710 | * dentry->d_lock must be held, returns with it unlocked. | |
711 | * Returns dentry requiring refcount drop, or NULL if we're done. | |
712 | */ | |
713 | static struct dentry *dentry_kill(struct dentry *dentry) | |
714 | __releases(dentry->d_lock) | |
715 | { | |
716 | struct inode *inode = dentry->d_inode; | |
717 | struct dentry *parent = NULL; | |
718 | ||
719 | if (inode && unlikely(!spin_trylock(&inode->i_lock))) | |
f657a666 | 720 | goto slow_positive; |
c1d0c1a2 JO |
721 | |
722 | if (!IS_ROOT(dentry)) { | |
723 | parent = dentry->d_parent; | |
724 | if (unlikely(!spin_trylock(&parent->d_lock))) { | |
f657a666 AV |
725 | parent = __lock_parent(dentry); |
726 | if (likely(inode || !dentry->d_inode)) | |
727 | goto got_locks; | |
728 | /* negative that became positive */ | |
729 | if (parent) | |
730 | spin_unlock(&parent->d_lock); | |
731 | inode = dentry->d_inode; | |
732 | goto slow_positive; | |
c1d0c1a2 JO |
733 | } |
734 | } | |
c1d0c1a2 JO |
735 | __dentry_kill(dentry); |
736 | return parent; | |
737 | ||
f657a666 AV |
738 | slow_positive: |
739 | spin_unlock(&dentry->d_lock); | |
740 | spin_lock(&inode->i_lock); | |
741 | spin_lock(&dentry->d_lock); | |
742 | parent = lock_parent(dentry); | |
743 | got_locks: | |
744 | if (unlikely(dentry->d_lockref.count != 1)) { | |
745 | dentry->d_lockref.count--; | |
746 | } else if (likely(!retain_dentry(dentry))) { | |
747 | __dentry_kill(dentry); | |
748 | return parent; | |
749 | } | |
750 | /* we are keeping it, after all */ | |
751 | if (inode) | |
752 | spin_unlock(&inode->i_lock); | |
753 | if (parent) | |
754 | spin_unlock(&parent->d_lock); | |
c1d0c1a2 | 755 | spin_unlock(&dentry->d_lock); |
f657a666 | 756 | return NULL; |
c1d0c1a2 JO |
757 | } |
758 | ||
360f5479 LT |
759 | /* |
760 | * Try to do a lockless dput(), and return whether that was successful. | |
761 | * | |
762 | * If unsuccessful, we return false, having already taken the dentry lock. | |
763 | * | |
764 | * The caller needs to hold the RCU read lock, so that the dentry is | |
765 | * guaranteed to stay around even if the refcount goes down to zero! | |
766 | */ | |
767 | static inline bool fast_dput(struct dentry *dentry) | |
768 | { | |
769 | int ret; | |
770 | unsigned int d_flags; | |
771 | ||
772 | /* | |
773 | * If we have a d_op->d_delete() operation, we sould not | |
75a6f82a | 774 | * let the dentry count go to zero, so use "put_or_lock". |
360f5479 LT |
775 | */ |
776 | if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) | |
777 | return lockref_put_or_lock(&dentry->d_lockref); | |
778 | ||
779 | /* | |
780 | * .. otherwise, we can try to just decrement the | |
781 | * lockref optimistically. | |
782 | */ | |
783 | ret = lockref_put_return(&dentry->d_lockref); | |
784 | ||
785 | /* | |
786 | * If the lockref_put_return() failed due to the lock being held | |
787 | * by somebody else, the fast path has failed. We will need to | |
788 | * get the lock, and then check the count again. | |
789 | */ | |
790 | if (unlikely(ret < 0)) { | |
791 | spin_lock(&dentry->d_lock); | |
792 | if (dentry->d_lockref.count > 1) { | |
793 | dentry->d_lockref.count--; | |
794 | spin_unlock(&dentry->d_lock); | |
7964410f | 795 | return true; |
360f5479 | 796 | } |
7964410f | 797 | return false; |
360f5479 LT |
798 | } |
799 | ||
800 | /* | |
801 | * If we weren't the last ref, we're done. | |
802 | */ | |
803 | if (ret) | |
7964410f | 804 | return true; |
360f5479 LT |
805 | |
806 | /* | |
807 | * Careful, careful. The reference count went down | |
808 | * to zero, but we don't hold the dentry lock, so | |
809 | * somebody else could get it again, and do another | |
810 | * dput(), and we need to not race with that. | |
811 | * | |
812 | * However, there is a very special and common case | |
813 | * where we don't care, because there is nothing to | |
814 | * do: the dentry is still hashed, it does not have | |
815 | * a 'delete' op, and it's referenced and already on | |
816 | * the LRU list. | |
817 | * | |
818 | * NOTE! Since we aren't locked, these values are | |
819 | * not "stable". However, it is sufficient that at | |
820 | * some point after we dropped the reference the | |
821 | * dentry was hashed and the flags had the proper | |
822 | * value. Other dentry users may have re-gotten | |
823 | * a reference to the dentry and change that, but | |
824 | * our work is done - we can leave the dentry | |
825 | * around with a zero refcount. | |
77573fa3 HL |
826 | * |
827 | * Nevertheless, there are two cases that we should kill | |
828 | * the dentry anyway. | |
829 | * 1. free disconnected dentries as soon as their refcount | |
830 | * reached zero. | |
831 | * 2. free dentries if they should not be cached. | |
360f5479 LT |
832 | */ |
833 | smp_rmb(); | |
66702eb5 | 834 | d_flags = READ_ONCE(dentry->d_flags); |
77573fa3 HL |
835 | d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | |
836 | DCACHE_DISCONNECTED | DCACHE_DONTCACHE; | |
360f5479 LT |
837 | |
838 | /* Nothing to do? Dropping the reference was all we needed? */ | |
839 | if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry)) | |
7964410f | 840 | return true; |
360f5479 LT |
841 | |
842 | /* | |
843 | * Not the fast normal case? Get the lock. We've already decremented | |
844 | * the refcount, but we'll need to re-check the situation after | |
845 | * getting the lock. | |
846 | */ | |
847 | spin_lock(&dentry->d_lock); | |
848 | ||
849 | /* | |
850 | * Did somebody else grab a reference to it in the meantime, and | |
851 | * we're no longer the last user after all? Alternatively, somebody | |
852 | * else could have killed it and marked it dead. Either way, we | |
853 | * don't need to do anything else. | |
854 | */ | |
855 | if (dentry->d_lockref.count) { | |
856 | spin_unlock(&dentry->d_lock); | |
7964410f | 857 | return true; |
360f5479 LT |
858 | } |
859 | ||
860 | /* | |
861 | * Re-get the reference we optimistically dropped. We hold the | |
862 | * lock, and we just tested that it was zero, so we can just | |
863 | * set it to 1. | |
864 | */ | |
865 | dentry->d_lockref.count = 1; | |
7964410f | 866 | return false; |
360f5479 LT |
867 | } |
868 | ||
869 | ||
1da177e4 LT |
870 | /* |
871 | * This is dput | |
872 | * | |
873 | * This is complicated by the fact that we do not want to put | |
874 | * dentries that are no longer on any hash chain on the unused | |
875 | * list: we'd much rather just get rid of them immediately. | |
876 | * | |
877 | * However, that implies that we have to traverse the dentry | |
878 | * tree upwards to the parents which might _also_ now be | |
879 | * scheduled for deletion (it may have been only waiting for | |
880 | * its last child to go away). | |
881 | * | |
882 | * This tail recursion is done by hand as we don't want to depend | |
883 | * on the compiler to always get this right (gcc generally doesn't). | |
884 | * Real recursion would eat up our stack space. | |
885 | */ | |
886 | ||
887 | /* | |
888 | * dput - release a dentry | |
889 | * @dentry: dentry to release | |
890 | * | |
891 | * Release a dentry. This will drop the usage count and if appropriate | |
892 | * call the dentry unlink method as well as removing it from the queues and | |
893 | * releasing its resources. If the parent dentries were scheduled for release | |
894 | * they too may now get deleted. | |
1da177e4 | 895 | */ |
1da177e4 LT |
896 | void dput(struct dentry *dentry) |
897 | { | |
1088a640 AV |
898 | while (dentry) { |
899 | might_sleep(); | |
1da177e4 | 900 | |
1088a640 AV |
901 | rcu_read_lock(); |
902 | if (likely(fast_dput(dentry))) { | |
903 | rcu_read_unlock(); | |
904 | return; | |
905 | } | |
47be6184 | 906 | |
1088a640 | 907 | /* Slow case: now with the dentry lock held */ |
360f5479 | 908 | rcu_read_unlock(); |
360f5479 | 909 | |
1088a640 AV |
910 | if (likely(retain_dentry(dentry))) { |
911 | spin_unlock(&dentry->d_lock); | |
912 | return; | |
913 | } | |
265ac902 | 914 | |
1088a640 | 915 | dentry = dentry_kill(dentry); |
47be6184 | 916 | } |
1da177e4 | 917 | } |
ec4f8605 | 918 | EXPORT_SYMBOL(dput); |
1da177e4 | 919 | |
9bdebc2b AV |
920 | static void __dput_to_list(struct dentry *dentry, struct list_head *list) |
921 | __must_hold(&dentry->d_lock) | |
922 | { | |
923 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { | |
924 | /* let the owner of the list it's on deal with it */ | |
925 | --dentry->d_lockref.count; | |
926 | } else { | |
927 | if (dentry->d_flags & DCACHE_LRU_LIST) | |
928 | d_lru_del(dentry); | |
929 | if (!--dentry->d_lockref.count) | |
930 | d_shrink_add(dentry, list); | |
931 | } | |
932 | } | |
933 | ||
934 | void dput_to_list(struct dentry *dentry, struct list_head *list) | |
935 | { | |
936 | rcu_read_lock(); | |
937 | if (likely(fast_dput(dentry))) { | |
938 | rcu_read_unlock(); | |
939 | return; | |
940 | } | |
941 | rcu_read_unlock(); | |
942 | if (!retain_dentry(dentry)) | |
943 | __dput_to_list(dentry, list); | |
944 | spin_unlock(&dentry->d_lock); | |
945 | } | |
1da177e4 | 946 | |
b5c84bf6 | 947 | /* This must be called with d_lock held */ |
dc0474be | 948 | static inline void __dget_dlock(struct dentry *dentry) |
23044507 | 949 | { |
98474236 | 950 | dentry->d_lockref.count++; |
23044507 NP |
951 | } |
952 | ||
dc0474be | 953 | static inline void __dget(struct dentry *dentry) |
1da177e4 | 954 | { |
98474236 | 955 | lockref_get(&dentry->d_lockref); |
1da177e4 LT |
956 | } |
957 | ||
b7ab39f6 NP |
958 | struct dentry *dget_parent(struct dentry *dentry) |
959 | { | |
df3d0bbc | 960 | int gotref; |
b7ab39f6 | 961 | struct dentry *ret; |
e8400933 | 962 | unsigned seq; |
b7ab39f6 | 963 | |
df3d0bbc WL |
964 | /* |
965 | * Do optimistic parent lookup without any | |
966 | * locking. | |
967 | */ | |
968 | rcu_read_lock(); | |
e8400933 | 969 | seq = raw_seqcount_begin(&dentry->d_seq); |
66702eb5 | 970 | ret = READ_ONCE(dentry->d_parent); |
df3d0bbc WL |
971 | gotref = lockref_get_not_zero(&ret->d_lockref); |
972 | rcu_read_unlock(); | |
973 | if (likely(gotref)) { | |
e8400933 | 974 | if (!read_seqcount_retry(&dentry->d_seq, seq)) |
df3d0bbc WL |
975 | return ret; |
976 | dput(ret); | |
977 | } | |
978 | ||
b7ab39f6 | 979 | repeat: |
a734eb45 NP |
980 | /* |
981 | * Don't need rcu_dereference because we re-check it was correct under | |
982 | * the lock. | |
983 | */ | |
984 | rcu_read_lock(); | |
b7ab39f6 | 985 | ret = dentry->d_parent; |
a734eb45 NP |
986 | spin_lock(&ret->d_lock); |
987 | if (unlikely(ret != dentry->d_parent)) { | |
988 | spin_unlock(&ret->d_lock); | |
989 | rcu_read_unlock(); | |
b7ab39f6 NP |
990 | goto repeat; |
991 | } | |
a734eb45 | 992 | rcu_read_unlock(); |
98474236 WL |
993 | BUG_ON(!ret->d_lockref.count); |
994 | ret->d_lockref.count++; | |
b7ab39f6 | 995 | spin_unlock(&ret->d_lock); |
b7ab39f6 NP |
996 | return ret; |
997 | } | |
998 | EXPORT_SYMBOL(dget_parent); | |
999 | ||
61fec493 AV |
1000 | static struct dentry * __d_find_any_alias(struct inode *inode) |
1001 | { | |
1002 | struct dentry *alias; | |
1003 | ||
1004 | if (hlist_empty(&inode->i_dentry)) | |
1005 | return NULL; | |
1006 | alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias); | |
1007 | __dget(alias); | |
1008 | return alias; | |
1009 | } | |
1010 | ||
1011 | /** | |
1012 | * d_find_any_alias - find any alias for a given inode | |
1013 | * @inode: inode to find an alias for | |
1014 | * | |
1015 | * If any aliases exist for the given inode, take and return a | |
1016 | * reference for one of them. If no aliases exist, return %NULL. | |
1017 | */ | |
1018 | struct dentry *d_find_any_alias(struct inode *inode) | |
1019 | { | |
1020 | struct dentry *de; | |
1021 | ||
1022 | spin_lock(&inode->i_lock); | |
1023 | de = __d_find_any_alias(inode); | |
1024 | spin_unlock(&inode->i_lock); | |
1025 | return de; | |
1026 | } | |
1027 | EXPORT_SYMBOL(d_find_any_alias); | |
1028 | ||
52ed46f0 | 1029 | static struct dentry *__d_find_alias(struct inode *inode) |
1da177e4 | 1030 | { |
61fec493 AV |
1031 | struct dentry *alias; |
1032 | ||
1033 | if (S_ISDIR(inode->i_mode)) | |
1034 | return __d_find_any_alias(inode); | |
1da177e4 | 1035 | |
946e51f2 | 1036 | hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { |
da502956 | 1037 | spin_lock(&alias->d_lock); |
61fec493 | 1038 | if (!d_unhashed(alias)) { |
8d80d7da BF |
1039 | __dget_dlock(alias); |
1040 | spin_unlock(&alias->d_lock); | |
1041 | return alias; | |
1da177e4 | 1042 | } |
da502956 | 1043 | spin_unlock(&alias->d_lock); |
1da177e4 | 1044 | } |
da502956 | 1045 | return NULL; |
1da177e4 LT |
1046 | } |
1047 | ||
961f3c89 MCC |
1048 | /** |
1049 | * d_find_alias - grab a hashed alias of inode | |
1050 | * @inode: inode in question | |
1051 | * | |
1052 | * If inode has a hashed alias, or is a directory and has any alias, | |
1053 | * acquire the reference to alias and return it. Otherwise return NULL. | |
1054 | * Notice that if inode is a directory there can be only one alias and | |
1055 | * it can be unhashed only if it has no children, or if it is the root | |
1056 | * of a filesystem, or if the directory was renamed and d_revalidate | |
1057 | * was the first vfs operation to notice. | |
1058 | * | |
1059 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer | |
1060 | * any other hashed alias over that one. | |
1061 | */ | |
da502956 | 1062 | struct dentry *d_find_alias(struct inode *inode) |
1da177e4 | 1063 | { |
214fda1f DH |
1064 | struct dentry *de = NULL; |
1065 | ||
b3d9b7a3 | 1066 | if (!hlist_empty(&inode->i_dentry)) { |
873feea0 | 1067 | spin_lock(&inode->i_lock); |
52ed46f0 | 1068 | de = __d_find_alias(inode); |
873feea0 | 1069 | spin_unlock(&inode->i_lock); |
214fda1f | 1070 | } |
1da177e4 LT |
1071 | return de; |
1072 | } | |
ec4f8605 | 1073 | EXPORT_SYMBOL(d_find_alias); |
1da177e4 | 1074 | |
bca585d2 AV |
1075 | /* |
1076 | * Caller MUST be holding rcu_read_lock() and be guaranteed | |
1077 | * that inode won't get freed until rcu_read_unlock(). | |
1078 | */ | |
1079 | struct dentry *d_find_alias_rcu(struct inode *inode) | |
1080 | { | |
1081 | struct hlist_head *l = &inode->i_dentry; | |
1082 | struct dentry *de = NULL; | |
1083 | ||
1084 | spin_lock(&inode->i_lock); | |
1085 | // ->i_dentry and ->i_rcu are colocated, but the latter won't be | |
1086 | // used without having I_FREEING set, which means no aliases left | |
1087 | if (likely(!(inode->i_state & I_FREEING) && !hlist_empty(l))) { | |
1088 | if (S_ISDIR(inode->i_mode)) { | |
1089 | de = hlist_entry(l->first, struct dentry, d_u.d_alias); | |
1090 | } else { | |
1091 | hlist_for_each_entry(de, l, d_u.d_alias) | |
1092 | if (!d_unhashed(de)) | |
1093 | break; | |
1094 | } | |
1095 | } | |
1096 | spin_unlock(&inode->i_lock); | |
1097 | return de; | |
1098 | } | |
1099 | ||
1da177e4 LT |
1100 | /* |
1101 | * Try to kill dentries associated with this inode. | |
1102 | * WARNING: you must own a reference to inode. | |
1103 | */ | |
1104 | void d_prune_aliases(struct inode *inode) | |
1105 | { | |
0cdca3f9 | 1106 | struct dentry *dentry; |
1da177e4 | 1107 | restart: |
873feea0 | 1108 | spin_lock(&inode->i_lock); |
946e51f2 | 1109 | hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) { |
1da177e4 | 1110 | spin_lock(&dentry->d_lock); |
98474236 | 1111 | if (!dentry->d_lockref.count) { |
29355c39 AV |
1112 | struct dentry *parent = lock_parent(dentry); |
1113 | if (likely(!dentry->d_lockref.count)) { | |
1114 | __dentry_kill(dentry); | |
4a7795d3 | 1115 | dput(parent); |
29355c39 AV |
1116 | goto restart; |
1117 | } | |
1118 | if (parent) | |
1119 | spin_unlock(&parent->d_lock); | |
1da177e4 LT |
1120 | } |
1121 | spin_unlock(&dentry->d_lock); | |
1122 | } | |
873feea0 | 1123 | spin_unlock(&inode->i_lock); |
1da177e4 | 1124 | } |
ec4f8605 | 1125 | EXPORT_SYMBOL(d_prune_aliases); |
1da177e4 | 1126 | |
3b3f09f4 AV |
1127 | /* |
1128 | * Lock a dentry from shrink list. | |
8f04da2a JO |
1129 | * Called under rcu_read_lock() and dentry->d_lock; the former |
1130 | * guarantees that nothing we access will be freed under us. | |
3b3f09f4 | 1131 | * Note that dentry is *not* protected from concurrent dentry_kill(), |
8f04da2a JO |
1132 | * d_delete(), etc. |
1133 | * | |
3b3f09f4 AV |
1134 | * Return false if dentry has been disrupted or grabbed, leaving |
1135 | * the caller to kick it off-list. Otherwise, return true and have | |
1136 | * that dentry's inode and parent both locked. | |
1137 | */ | |
1138 | static bool shrink_lock_dentry(struct dentry *dentry) | |
1da177e4 | 1139 | { |
3b3f09f4 AV |
1140 | struct inode *inode; |
1141 | struct dentry *parent; | |
da3bbdd4 | 1142 | |
3b3f09f4 AV |
1143 | if (dentry->d_lockref.count) |
1144 | return false; | |
1145 | ||
1146 | inode = dentry->d_inode; | |
1147 | if (inode && unlikely(!spin_trylock(&inode->i_lock))) { | |
3b3f09f4 AV |
1148 | spin_unlock(&dentry->d_lock); |
1149 | spin_lock(&inode->i_lock); | |
ec33679d | 1150 | spin_lock(&dentry->d_lock); |
3b3f09f4 AV |
1151 | if (unlikely(dentry->d_lockref.count)) |
1152 | goto out; | |
1153 | /* changed inode means that somebody had grabbed it */ | |
1154 | if (unlikely(inode != dentry->d_inode)) | |
1155 | goto out; | |
3b3f09f4 | 1156 | } |
046b961b | 1157 | |
3b3f09f4 AV |
1158 | parent = dentry->d_parent; |
1159 | if (IS_ROOT(dentry) || likely(spin_trylock(&parent->d_lock))) | |
1160 | return true; | |
dd1f6b2e | 1161 | |
3b3f09f4 | 1162 | spin_unlock(&dentry->d_lock); |
3b3f09f4 AV |
1163 | spin_lock(&parent->d_lock); |
1164 | if (unlikely(parent != dentry->d_parent)) { | |
1165 | spin_unlock(&parent->d_lock); | |
1166 | spin_lock(&dentry->d_lock); | |
1167 | goto out; | |
1168 | } | |
1169 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
8f04da2a | 1170 | if (likely(!dentry->d_lockref.count)) |
3b3f09f4 | 1171 | return true; |
3b3f09f4 AV |
1172 | spin_unlock(&parent->d_lock); |
1173 | out: | |
1174 | if (inode) | |
1175 | spin_unlock(&inode->i_lock); | |
3b3f09f4 AV |
1176 | return false; |
1177 | } | |
77812a1e | 1178 | |
9bdebc2b | 1179 | void shrink_dentry_list(struct list_head *list) |
3b3f09f4 AV |
1180 | { |
1181 | while (!list_empty(list)) { | |
1182 | struct dentry *dentry, *parent; | |
64fd72e0 | 1183 | |
3b3f09f4 AV |
1184 | dentry = list_entry(list->prev, struct dentry, d_lru); |
1185 | spin_lock(&dentry->d_lock); | |
8f04da2a | 1186 | rcu_read_lock(); |
3b3f09f4 AV |
1187 | if (!shrink_lock_dentry(dentry)) { |
1188 | bool can_free = false; | |
8f04da2a | 1189 | rcu_read_unlock(); |
3b3f09f4 AV |
1190 | d_shrink_del(dentry); |
1191 | if (dentry->d_lockref.count < 0) | |
1192 | can_free = dentry->d_flags & DCACHE_MAY_FREE; | |
64fd72e0 AV |
1193 | spin_unlock(&dentry->d_lock); |
1194 | if (can_free) | |
1195 | dentry_free(dentry); | |
1196 | continue; | |
1197 | } | |
8f04da2a | 1198 | rcu_read_unlock(); |
3b3f09f4 AV |
1199 | d_shrink_del(dentry); |
1200 | parent = dentry->d_parent; | |
9bdebc2b AV |
1201 | if (parent != dentry) |
1202 | __dput_to_list(parent, list); | |
ff2fde99 | 1203 | __dentry_kill(dentry); |
da3bbdd4 | 1204 | } |
3049cfe2 CH |
1205 | } |
1206 | ||
3f97b163 VD |
1207 | static enum lru_status dentry_lru_isolate(struct list_head *item, |
1208 | struct list_lru_one *lru, spinlock_t *lru_lock, void *arg) | |
f6041567 DC |
1209 | { |
1210 | struct list_head *freeable = arg; | |
1211 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
1212 | ||
1213 | ||
1214 | /* | |
1215 | * we are inverting the lru lock/dentry->d_lock here, | |
1216 | * so use a trylock. If we fail to get the lock, just skip | |
1217 | * it | |
1218 | */ | |
1219 | if (!spin_trylock(&dentry->d_lock)) | |
1220 | return LRU_SKIP; | |
1221 | ||
1222 | /* | |
1223 | * Referenced dentries are still in use. If they have active | |
1224 | * counts, just remove them from the LRU. Otherwise give them | |
1225 | * another pass through the LRU. | |
1226 | */ | |
1227 | if (dentry->d_lockref.count) { | |
3f97b163 | 1228 | d_lru_isolate(lru, dentry); |
f6041567 DC |
1229 | spin_unlock(&dentry->d_lock); |
1230 | return LRU_REMOVED; | |
1231 | } | |
1232 | ||
1233 | if (dentry->d_flags & DCACHE_REFERENCED) { | |
1234 | dentry->d_flags &= ~DCACHE_REFERENCED; | |
1235 | spin_unlock(&dentry->d_lock); | |
1236 | ||
1237 | /* | |
1238 | * The list move itself will be made by the common LRU code. At | |
1239 | * this point, we've dropped the dentry->d_lock but keep the | |
1240 | * lru lock. This is safe to do, since every list movement is | |
1241 | * protected by the lru lock even if both locks are held. | |
1242 | * | |
1243 | * This is guaranteed by the fact that all LRU management | |
1244 | * functions are intermediated by the LRU API calls like | |
0a97c01c | 1245 | * list_lru_add_obj and list_lru_del_obj. List movement in this file |
f6041567 DC |
1246 | * only ever occur through this functions or through callbacks |
1247 | * like this one, that are called from the LRU API. | |
1248 | * | |
1249 | * The only exceptions to this are functions like | |
1250 | * shrink_dentry_list, and code that first checks for the | |
1251 | * DCACHE_SHRINK_LIST flag. Those are guaranteed to be | |
1252 | * operating only with stack provided lists after they are | |
1253 | * properly isolated from the main list. It is thus, always a | |
1254 | * local access. | |
1255 | */ | |
1256 | return LRU_ROTATE; | |
1257 | } | |
1258 | ||
3f97b163 | 1259 | d_lru_shrink_move(lru, dentry, freeable); |
f6041567 DC |
1260 | spin_unlock(&dentry->d_lock); |
1261 | ||
1262 | return LRU_REMOVED; | |
1263 | } | |
1264 | ||
3049cfe2 | 1265 | /** |
b48f03b3 DC |
1266 | * prune_dcache_sb - shrink the dcache |
1267 | * @sb: superblock | |
503c358c | 1268 | * @sc: shrink control, passed to list_lru_shrink_walk() |
b48f03b3 | 1269 | * |
503c358c VD |
1270 | * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This |
1271 | * is done when we need more memory and called from the superblock shrinker | |
b48f03b3 | 1272 | * function. |
3049cfe2 | 1273 | * |
b48f03b3 DC |
1274 | * This function may fail to free any resources if all the dentries are in |
1275 | * use. | |
3049cfe2 | 1276 | */ |
503c358c | 1277 | long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc) |
3049cfe2 | 1278 | { |
f6041567 DC |
1279 | LIST_HEAD(dispose); |
1280 | long freed; | |
3049cfe2 | 1281 | |
503c358c VD |
1282 | freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc, |
1283 | dentry_lru_isolate, &dispose); | |
f6041567 | 1284 | shrink_dentry_list(&dispose); |
0a234c6d | 1285 | return freed; |
da3bbdd4 | 1286 | } |
23044507 | 1287 | |
4e717f5c | 1288 | static enum lru_status dentry_lru_isolate_shrink(struct list_head *item, |
3f97b163 | 1289 | struct list_lru_one *lru, spinlock_t *lru_lock, void *arg) |
dd1f6b2e | 1290 | { |
4e717f5c GC |
1291 | struct list_head *freeable = arg; |
1292 | struct dentry *dentry = container_of(item, struct dentry, d_lru); | |
dd1f6b2e | 1293 | |
4e717f5c GC |
1294 | /* |
1295 | * we are inverting the lru lock/dentry->d_lock here, | |
1296 | * so use a trylock. If we fail to get the lock, just skip | |
1297 | * it | |
1298 | */ | |
1299 | if (!spin_trylock(&dentry->d_lock)) | |
1300 | return LRU_SKIP; | |
1301 | ||
3f97b163 | 1302 | d_lru_shrink_move(lru, dentry, freeable); |
4e717f5c | 1303 | spin_unlock(&dentry->d_lock); |
ec33679d | 1304 | |
4e717f5c | 1305 | return LRU_REMOVED; |
da3bbdd4 KM |
1306 | } |
1307 | ||
4e717f5c | 1308 | |
1da177e4 LT |
1309 | /** |
1310 | * shrink_dcache_sb - shrink dcache for a superblock | |
1311 | * @sb: superblock | |
1312 | * | |
3049cfe2 CH |
1313 | * Shrink the dcache for the specified super block. This is used to free |
1314 | * the dcache before unmounting a file system. | |
1da177e4 | 1315 | */ |
3049cfe2 | 1316 | void shrink_dcache_sb(struct super_block *sb) |
1da177e4 | 1317 | { |
4e717f5c GC |
1318 | do { |
1319 | LIST_HEAD(dispose); | |
1320 | ||
1dbd449c | 1321 | list_lru_walk(&sb->s_dentry_lru, |
b17c070f | 1322 | dentry_lru_isolate_shrink, &dispose, 1024); |
4e717f5c | 1323 | shrink_dentry_list(&dispose); |
b17c070f | 1324 | } while (list_lru_count(&sb->s_dentry_lru) > 0); |
1da177e4 | 1325 | } |
ec4f8605 | 1326 | EXPORT_SYMBOL(shrink_dcache_sb); |
1da177e4 | 1327 | |
db14fc3a MS |
1328 | /** |
1329 | * enum d_walk_ret - action to talke during tree walk | |
1330 | * @D_WALK_CONTINUE: contrinue walk | |
1331 | * @D_WALK_QUIT: quit walk | |
1332 | * @D_WALK_NORETRY: quit when retry is needed | |
1333 | * @D_WALK_SKIP: skip this dentry and its children | |
1334 | */ | |
1335 | enum d_walk_ret { | |
1336 | D_WALK_CONTINUE, | |
1337 | D_WALK_QUIT, | |
1338 | D_WALK_NORETRY, | |
1339 | D_WALK_SKIP, | |
1340 | }; | |
c826cb7d | 1341 | |
1da177e4 | 1342 | /** |
db14fc3a MS |
1343 | * d_walk - walk the dentry tree |
1344 | * @parent: start of walk | |
1345 | * @data: data passed to @enter() and @finish() | |
1346 | * @enter: callback when first entering the dentry | |
1da177e4 | 1347 | * |
3a8e3611 | 1348 | * The @enter() callbacks are called with d_lock held. |
1da177e4 | 1349 | */ |
db14fc3a | 1350 | static void d_walk(struct dentry *parent, void *data, |
3a8e3611 | 1351 | enum d_walk_ret (*enter)(void *, struct dentry *)) |
1da177e4 | 1352 | { |
949854d0 | 1353 | struct dentry *this_parent; |
1da177e4 | 1354 | struct list_head *next; |
48f5ec21 | 1355 | unsigned seq = 0; |
db14fc3a MS |
1356 | enum d_walk_ret ret; |
1357 | bool retry = true; | |
949854d0 | 1358 | |
58db63d0 | 1359 | again: |
48f5ec21 | 1360 | read_seqbegin_or_lock(&rename_lock, &seq); |
58db63d0 | 1361 | this_parent = parent; |
2fd6b7f5 | 1362 | spin_lock(&this_parent->d_lock); |
db14fc3a MS |
1363 | |
1364 | ret = enter(data, this_parent); | |
1365 | switch (ret) { | |
1366 | case D_WALK_CONTINUE: | |
1367 | break; | |
1368 | case D_WALK_QUIT: | |
1369 | case D_WALK_SKIP: | |
1370 | goto out_unlock; | |
1371 | case D_WALK_NORETRY: | |
1372 | retry = false; | |
1373 | break; | |
1374 | } | |
1da177e4 LT |
1375 | repeat: |
1376 | next = this_parent->d_subdirs.next; | |
1377 | resume: | |
1378 | while (next != &this_parent->d_subdirs) { | |
1379 | struct list_head *tmp = next; | |
946e51f2 | 1380 | struct dentry *dentry = list_entry(tmp, struct dentry, d_child); |
1da177e4 | 1381 | next = tmp->next; |
2fd6b7f5 | 1382 | |
ba65dc5e AV |
1383 | if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR)) |
1384 | continue; | |
1385 | ||
2fd6b7f5 | 1386 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
db14fc3a MS |
1387 | |
1388 | ret = enter(data, dentry); | |
1389 | switch (ret) { | |
1390 | case D_WALK_CONTINUE: | |
1391 | break; | |
1392 | case D_WALK_QUIT: | |
2fd6b7f5 | 1393 | spin_unlock(&dentry->d_lock); |
db14fc3a MS |
1394 | goto out_unlock; |
1395 | case D_WALK_NORETRY: | |
1396 | retry = false; | |
1397 | break; | |
1398 | case D_WALK_SKIP: | |
1399 | spin_unlock(&dentry->d_lock); | |
1400 | continue; | |
2fd6b7f5 | 1401 | } |
db14fc3a | 1402 | |
1da177e4 | 1403 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 | 1404 | spin_unlock(&this_parent->d_lock); |
5facae4f | 1405 | spin_release(&dentry->d_lock.dep_map, _RET_IP_); |
1da177e4 | 1406 | this_parent = dentry; |
2fd6b7f5 | 1407 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1408 | goto repeat; |
1409 | } | |
2fd6b7f5 | 1410 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
1411 | } |
1412 | /* | |
1413 | * All done at this level ... ascend and resume the search. | |
1414 | */ | |
ca5358ef AV |
1415 | rcu_read_lock(); |
1416 | ascend: | |
1da177e4 | 1417 | if (this_parent != parent) { |
c826cb7d | 1418 | struct dentry *child = this_parent; |
31dec132 AV |
1419 | this_parent = child->d_parent; |
1420 | ||
31dec132 AV |
1421 | spin_unlock(&child->d_lock); |
1422 | spin_lock(&this_parent->d_lock); | |
1423 | ||
ca5358ef AV |
1424 | /* might go back up the wrong parent if we have had a rename. */ |
1425 | if (need_seqretry(&rename_lock, seq)) | |
949854d0 | 1426 | goto rename_retry; |
2159184e AV |
1427 | /* go into the first sibling still alive */ |
1428 | do { | |
1429 | next = child->d_child.next; | |
ca5358ef AV |
1430 | if (next == &this_parent->d_subdirs) |
1431 | goto ascend; | |
1432 | child = list_entry(next, struct dentry, d_child); | |
2159184e | 1433 | } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED)); |
31dec132 | 1434 | rcu_read_unlock(); |
1da177e4 LT |
1435 | goto resume; |
1436 | } | |
ca5358ef | 1437 | if (need_seqretry(&rename_lock, seq)) |
949854d0 | 1438 | goto rename_retry; |
ca5358ef | 1439 | rcu_read_unlock(); |
db14fc3a MS |
1440 | |
1441 | out_unlock: | |
1442 | spin_unlock(&this_parent->d_lock); | |
48f5ec21 | 1443 | done_seqretry(&rename_lock, seq); |
db14fc3a | 1444 | return; |
58db63d0 NP |
1445 | |
1446 | rename_retry: | |
ca5358ef AV |
1447 | spin_unlock(&this_parent->d_lock); |
1448 | rcu_read_unlock(); | |
1449 | BUG_ON(seq & 1); | |
db14fc3a MS |
1450 | if (!retry) |
1451 | return; | |
48f5ec21 | 1452 | seq = 1; |
58db63d0 | 1453 | goto again; |
1da177e4 | 1454 | } |
db14fc3a | 1455 | |
01619491 IK |
1456 | struct check_mount { |
1457 | struct vfsmount *mnt; | |
1458 | unsigned int mounted; | |
1459 | }; | |
1460 | ||
1461 | static enum d_walk_ret path_check_mount(void *data, struct dentry *dentry) | |
1462 | { | |
1463 | struct check_mount *info = data; | |
1464 | struct path path = { .mnt = info->mnt, .dentry = dentry }; | |
1465 | ||
1466 | if (likely(!d_mountpoint(dentry))) | |
1467 | return D_WALK_CONTINUE; | |
1468 | if (__path_is_mountpoint(&path)) { | |
1469 | info->mounted = 1; | |
1470 | return D_WALK_QUIT; | |
1471 | } | |
1472 | return D_WALK_CONTINUE; | |
1473 | } | |
1474 | ||
1475 | /** | |
1476 | * path_has_submounts - check for mounts over a dentry in the | |
1477 | * current namespace. | |
1478 | * @parent: path to check. | |
1479 | * | |
1480 | * Return true if the parent or its subdirectories contain | |
1481 | * a mount point in the current namespace. | |
1482 | */ | |
1483 | int path_has_submounts(const struct path *parent) | |
1484 | { | |
1485 | struct check_mount data = { .mnt = parent->mnt, .mounted = 0 }; | |
1486 | ||
1487 | read_seqlock_excl(&mount_lock); | |
3a8e3611 | 1488 | d_walk(parent->dentry, &data, path_check_mount); |
01619491 IK |
1489 | read_sequnlock_excl(&mount_lock); |
1490 | ||
1491 | return data.mounted; | |
1492 | } | |
1493 | EXPORT_SYMBOL(path_has_submounts); | |
1494 | ||
eed81007 MS |
1495 | /* |
1496 | * Called by mount code to set a mountpoint and check if the mountpoint is | |
1497 | * reachable (e.g. NFS can unhash a directory dentry and then the complete | |
1498 | * subtree can become unreachable). | |
1499 | * | |
1ffe46d1 | 1500 | * Only one of d_invalidate() and d_set_mounted() must succeed. For |
eed81007 MS |
1501 | * this reason take rename_lock and d_lock on dentry and ancestors. |
1502 | */ | |
1503 | int d_set_mounted(struct dentry *dentry) | |
1504 | { | |
1505 | struct dentry *p; | |
1506 | int ret = -ENOENT; | |
1507 | write_seqlock(&rename_lock); | |
1508 | for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) { | |
1ffe46d1 | 1509 | /* Need exclusion wrt. d_invalidate() */ |
eed81007 MS |
1510 | spin_lock(&p->d_lock); |
1511 | if (unlikely(d_unhashed(p))) { | |
1512 | spin_unlock(&p->d_lock); | |
1513 | goto out; | |
1514 | } | |
1515 | spin_unlock(&p->d_lock); | |
1516 | } | |
1517 | spin_lock(&dentry->d_lock); | |
1518 | if (!d_unlinked(dentry)) { | |
3895dbf8 EB |
1519 | ret = -EBUSY; |
1520 | if (!d_mountpoint(dentry)) { | |
1521 | dentry->d_flags |= DCACHE_MOUNTED; | |
1522 | ret = 0; | |
1523 | } | |
eed81007 MS |
1524 | } |
1525 | spin_unlock(&dentry->d_lock); | |
1526 | out: | |
1527 | write_sequnlock(&rename_lock); | |
1528 | return ret; | |
1529 | } | |
1530 | ||
1da177e4 | 1531 | /* |
fd517909 | 1532 | * Search the dentry child list of the specified parent, |
1da177e4 LT |
1533 | * and move any unused dentries to the end of the unused |
1534 | * list for prune_dcache(). We descend to the next level | |
1535 | * whenever the d_subdirs list is non-empty and continue | |
1536 | * searching. | |
1537 | * | |
1538 | * It returns zero iff there are no unused children, | |
1539 | * otherwise it returns the number of children moved to | |
1540 | * the end of the unused list. This may not be the total | |
1541 | * number of unused children, because select_parent can | |
1542 | * drop the lock and return early due to latency | |
1543 | * constraints. | |
1544 | */ | |
1da177e4 | 1545 | |
db14fc3a MS |
1546 | struct select_data { |
1547 | struct dentry *start; | |
9bdebc2b AV |
1548 | union { |
1549 | long found; | |
1550 | struct dentry *victim; | |
1551 | }; | |
db14fc3a | 1552 | struct list_head dispose; |
db14fc3a | 1553 | }; |
23044507 | 1554 | |
db14fc3a MS |
1555 | static enum d_walk_ret select_collect(void *_data, struct dentry *dentry) |
1556 | { | |
1557 | struct select_data *data = _data; | |
1558 | enum d_walk_ret ret = D_WALK_CONTINUE; | |
1da177e4 | 1559 | |
db14fc3a MS |
1560 | if (data->start == dentry) |
1561 | goto out; | |
2fd6b7f5 | 1562 | |
fe91522a | 1563 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { |
db14fc3a | 1564 | data->found++; |
fe91522a AV |
1565 | } else { |
1566 | if (dentry->d_flags & DCACHE_LRU_LIST) | |
1567 | d_lru_del(dentry); | |
1568 | if (!dentry->d_lockref.count) { | |
1569 | d_shrink_add(dentry, &data->dispose); | |
1570 | data->found++; | |
1571 | } | |
1da177e4 | 1572 | } |
db14fc3a MS |
1573 | /* |
1574 | * We can return to the caller if we have found some (this | |
1575 | * ensures forward progress). We'll be coming back to find | |
1576 | * the rest. | |
1577 | */ | |
fe91522a AV |
1578 | if (!list_empty(&data->dispose)) |
1579 | ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY; | |
1da177e4 | 1580 | out: |
db14fc3a | 1581 | return ret; |
1da177e4 LT |
1582 | } |
1583 | ||
9bdebc2b AV |
1584 | static enum d_walk_ret select_collect2(void *_data, struct dentry *dentry) |
1585 | { | |
1586 | struct select_data *data = _data; | |
1587 | enum d_walk_ret ret = D_WALK_CONTINUE; | |
1588 | ||
1589 | if (data->start == dentry) | |
1590 | goto out; | |
1591 | ||
1592 | if (dentry->d_flags & DCACHE_SHRINK_LIST) { | |
1593 | if (!dentry->d_lockref.count) { | |
1594 | rcu_read_lock(); | |
1595 | data->victim = dentry; | |
1596 | return D_WALK_QUIT; | |
1597 | } | |
1598 | } else { | |
1599 | if (dentry->d_flags & DCACHE_LRU_LIST) | |
1600 | d_lru_del(dentry); | |
1601 | if (!dentry->d_lockref.count) | |
1602 | d_shrink_add(dentry, &data->dispose); | |
1603 | } | |
1604 | /* | |
1605 | * We can return to the caller if we have found some (this | |
1606 | * ensures forward progress). We'll be coming back to find | |
1607 | * the rest. | |
1608 | */ | |
1609 | if (!list_empty(&data->dispose)) | |
1610 | ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY; | |
1611 | out: | |
1612 | return ret; | |
1613 | } | |
1614 | ||
1da177e4 LT |
1615 | /** |
1616 | * shrink_dcache_parent - prune dcache | |
1617 | * @parent: parent of entries to prune | |
1618 | * | |
1619 | * Prune the dcache to remove unused children of the parent dentry. | |
1620 | */ | |
db14fc3a | 1621 | void shrink_dcache_parent(struct dentry *parent) |
1da177e4 | 1622 | { |
db14fc3a | 1623 | for (;;) { |
9bdebc2b | 1624 | struct select_data data = {.start = parent}; |
1da177e4 | 1625 | |
db14fc3a | 1626 | INIT_LIST_HEAD(&data.dispose); |
3a8e3611 | 1627 | d_walk(parent, &data, select_collect); |
4fb48871 AV |
1628 | |
1629 | if (!list_empty(&data.dispose)) { | |
1630 | shrink_dentry_list(&data.dispose); | |
1631 | continue; | |
1632 | } | |
1633 | ||
1634 | cond_resched(); | |
db14fc3a MS |
1635 | if (!data.found) |
1636 | break; | |
9bdebc2b AV |
1637 | data.victim = NULL; |
1638 | d_walk(parent, &data, select_collect2); | |
1639 | if (data.victim) { | |
1640 | struct dentry *parent; | |
1641 | spin_lock(&data.victim->d_lock); | |
1642 | if (!shrink_lock_dentry(data.victim)) { | |
1643 | spin_unlock(&data.victim->d_lock); | |
1644 | rcu_read_unlock(); | |
1645 | } else { | |
1646 | rcu_read_unlock(); | |
1647 | parent = data.victim->d_parent; | |
1648 | if (parent != data.victim) | |
1649 | __dput_to_list(parent, &data.dispose); | |
1650 | __dentry_kill(data.victim); | |
1651 | } | |
1652 | } | |
1653 | if (!list_empty(&data.dispose)) | |
1654 | shrink_dentry_list(&data.dispose); | |
421348f1 | 1655 | } |
1da177e4 | 1656 | } |
ec4f8605 | 1657 | EXPORT_SYMBOL(shrink_dcache_parent); |
1da177e4 | 1658 | |
9c8c10e2 | 1659 | static enum d_walk_ret umount_check(void *_data, struct dentry *dentry) |
42c32608 | 1660 | { |
9c8c10e2 AV |
1661 | /* it has busy descendents; complain about those instead */ |
1662 | if (!list_empty(&dentry->d_subdirs)) | |
1663 | return D_WALK_CONTINUE; | |
42c32608 | 1664 | |
9c8c10e2 AV |
1665 | /* root with refcount 1 is fine */ |
1666 | if (dentry == _data && dentry->d_lockref.count == 1) | |
1667 | return D_WALK_CONTINUE; | |
1668 | ||
8c8e7dba | 1669 | WARN(1, "BUG: Dentry %p{i=%lx,n=%pd} " |
9c8c10e2 | 1670 | " still in use (%d) [unmount of %s %s]\n", |
42c32608 AV |
1671 | dentry, |
1672 | dentry->d_inode ? | |
1673 | dentry->d_inode->i_ino : 0UL, | |
9c8c10e2 | 1674 | dentry, |
42c32608 AV |
1675 | dentry->d_lockref.count, |
1676 | dentry->d_sb->s_type->name, | |
1677 | dentry->d_sb->s_id); | |
9c8c10e2 AV |
1678 | return D_WALK_CONTINUE; |
1679 | } | |
1680 | ||
1681 | static void do_one_tree(struct dentry *dentry) | |
1682 | { | |
1683 | shrink_dcache_parent(dentry); | |
3a8e3611 | 1684 | d_walk(dentry, dentry, umount_check); |
9c8c10e2 AV |
1685 | d_drop(dentry); |
1686 | dput(dentry); | |
42c32608 AV |
1687 | } |
1688 | ||
1689 | /* | |
1690 | * destroy the dentries attached to a superblock on unmounting | |
1691 | */ | |
1692 | void shrink_dcache_for_umount(struct super_block *sb) | |
1693 | { | |
1694 | struct dentry *dentry; | |
1695 | ||
9c8c10e2 | 1696 | WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked"); |
42c32608 AV |
1697 | |
1698 | dentry = sb->s_root; | |
1699 | sb->s_root = NULL; | |
9c8c10e2 | 1700 | do_one_tree(dentry); |
42c32608 | 1701 | |
f1ee6162 N |
1702 | while (!hlist_bl_empty(&sb->s_roots)) { |
1703 | dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_roots), struct dentry, d_hash)); | |
9c8c10e2 | 1704 | do_one_tree(dentry); |
42c32608 AV |
1705 | } |
1706 | } | |
1707 | ||
ff17fa56 | 1708 | static enum d_walk_ret find_submount(void *_data, struct dentry *dentry) |
848ac114 | 1709 | { |
ff17fa56 | 1710 | struct dentry **victim = _data; |
848ac114 | 1711 | if (d_mountpoint(dentry)) { |
8ed936b5 | 1712 | __dget_dlock(dentry); |
ff17fa56 | 1713 | *victim = dentry; |
848ac114 MS |
1714 | return D_WALK_QUIT; |
1715 | } | |
ff17fa56 | 1716 | return D_WALK_CONTINUE; |
848ac114 MS |
1717 | } |
1718 | ||
1719 | /** | |
1ffe46d1 EB |
1720 | * d_invalidate - detach submounts, prune dcache, and drop |
1721 | * @dentry: dentry to invalidate (aka detach, prune and drop) | |
848ac114 | 1722 | */ |
5542aa2f | 1723 | void d_invalidate(struct dentry *dentry) |
848ac114 | 1724 | { |
ff17fa56 | 1725 | bool had_submounts = false; |
1ffe46d1 EB |
1726 | spin_lock(&dentry->d_lock); |
1727 | if (d_unhashed(dentry)) { | |
1728 | spin_unlock(&dentry->d_lock); | |
5542aa2f | 1729 | return; |
1ffe46d1 | 1730 | } |
ff17fa56 | 1731 | __d_drop(dentry); |
1ffe46d1 EB |
1732 | spin_unlock(&dentry->d_lock); |
1733 | ||
848ac114 | 1734 | /* Negative dentries can be dropped without further checks */ |
ff17fa56 | 1735 | if (!dentry->d_inode) |
5542aa2f | 1736 | return; |
848ac114 | 1737 | |
ff17fa56 | 1738 | shrink_dcache_parent(dentry); |
848ac114 | 1739 | for (;;) { |
ff17fa56 | 1740 | struct dentry *victim = NULL; |
3a8e3611 | 1741 | d_walk(dentry, &victim, find_submount); |
ff17fa56 AV |
1742 | if (!victim) { |
1743 | if (had_submounts) | |
1744 | shrink_dcache_parent(dentry); | |
81be24d2 | 1745 | return; |
8ed936b5 | 1746 | } |
ff17fa56 AV |
1747 | had_submounts = true; |
1748 | detach_mounts(victim); | |
1749 | dput(victim); | |
848ac114 | 1750 | } |
848ac114 | 1751 | } |
1ffe46d1 | 1752 | EXPORT_SYMBOL(d_invalidate); |
848ac114 | 1753 | |
1da177e4 | 1754 | /** |
a4464dbc AV |
1755 | * __d_alloc - allocate a dcache entry |
1756 | * @sb: filesystem it will belong to | |
1da177e4 LT |
1757 | * @name: qstr of the name |
1758 | * | |
1759 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1760 | * available. On a success the dentry is returned. The name passed in is | |
1761 | * copied and the copy passed in may be reused after this call. | |
1762 | */ | |
1763 | ||
5c8b0dfc | 1764 | static struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name) |
1da177e4 LT |
1765 | { |
1766 | struct dentry *dentry; | |
1767 | char *dname; | |
285b102d | 1768 | int err; |
1da177e4 | 1769 | |
f53bf711 MS |
1770 | dentry = kmem_cache_alloc_lru(dentry_cache, &sb->s_dentry_lru, |
1771 | GFP_KERNEL); | |
1da177e4 LT |
1772 | if (!dentry) |
1773 | return NULL; | |
1774 | ||
6326c71f LT |
1775 | /* |
1776 | * We guarantee that the inline name is always NUL-terminated. | |
1777 | * This way the memcpy() done by the name switching in rename | |
1778 | * will still always have a NUL at the end, even if we might | |
1779 | * be overwriting an internal NUL character | |
1780 | */ | |
1781 | dentry->d_iname[DNAME_INLINE_LEN-1] = 0; | |
798434bd | 1782 | if (unlikely(!name)) { |
cdf01226 | 1783 | name = &slash_name; |
798434bd AV |
1784 | dname = dentry->d_iname; |
1785 | } else if (name->len > DNAME_INLINE_LEN-1) { | |
8d85b484 | 1786 | size_t size = offsetof(struct external_name, name[1]); |
2e03b4bc VB |
1787 | struct external_name *p = kmalloc(size + name->len, |
1788 | GFP_KERNEL_ACCOUNT | | |
1789 | __GFP_RECLAIMABLE); | |
1790 | if (!p) { | |
1da177e4 LT |
1791 | kmem_cache_free(dentry_cache, dentry); |
1792 | return NULL; | |
1793 | } | |
2e03b4bc VB |
1794 | atomic_set(&p->u.count, 1); |
1795 | dname = p->name; | |
1da177e4 LT |
1796 | } else { |
1797 | dname = dentry->d_iname; | |
1798 | } | |
1da177e4 LT |
1799 | |
1800 | dentry->d_name.len = name->len; | |
1801 | dentry->d_name.hash = name->hash; | |
1802 | memcpy(dname, name->name, name->len); | |
1803 | dname[name->len] = 0; | |
1804 | ||
6326c71f | 1805 | /* Make sure we always see the terminating NUL character */ |
7088efa9 | 1806 | smp_store_release(&dentry->d_name.name, dname); /* ^^^ */ |
6326c71f | 1807 | |
98474236 | 1808 | dentry->d_lockref.count = 1; |
dea3667b | 1809 | dentry->d_flags = 0; |
1da177e4 | 1810 | spin_lock_init(&dentry->d_lock); |
26475371 | 1811 | seqcount_spinlock_init(&dentry->d_seq, &dentry->d_lock); |
1da177e4 | 1812 | dentry->d_inode = NULL; |
a4464dbc AV |
1813 | dentry->d_parent = dentry; |
1814 | dentry->d_sb = sb; | |
1da177e4 LT |
1815 | dentry->d_op = NULL; |
1816 | dentry->d_fsdata = NULL; | |
ceb5bdc2 | 1817 | INIT_HLIST_BL_NODE(&dentry->d_hash); |
1da177e4 LT |
1818 | INIT_LIST_HEAD(&dentry->d_lru); |
1819 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
946e51f2 AV |
1820 | INIT_HLIST_NODE(&dentry->d_u.d_alias); |
1821 | INIT_LIST_HEAD(&dentry->d_child); | |
a4464dbc | 1822 | d_set_d_op(dentry, dentry->d_sb->s_d_op); |
1da177e4 | 1823 | |
285b102d MS |
1824 | if (dentry->d_op && dentry->d_op->d_init) { |
1825 | err = dentry->d_op->d_init(dentry); | |
1826 | if (err) { | |
1827 | if (dname_external(dentry)) | |
1828 | kfree(external_name(dentry)); | |
1829 | kmem_cache_free(dentry_cache, dentry); | |
1830 | return NULL; | |
1831 | } | |
1832 | } | |
1833 | ||
3e880fb5 | 1834 | this_cpu_inc(nr_dentry); |
312d3ca8 | 1835 | |
1da177e4 LT |
1836 | return dentry; |
1837 | } | |
a4464dbc AV |
1838 | |
1839 | /** | |
1840 | * d_alloc - allocate a dcache entry | |
1841 | * @parent: parent of entry to allocate | |
1842 | * @name: qstr of the name | |
1843 | * | |
1844 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1845 | * available. On a success the dentry is returned. The name passed in is | |
1846 | * copied and the copy passed in may be reused after this call. | |
1847 | */ | |
1848 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
1849 | { | |
1850 | struct dentry *dentry = __d_alloc(parent->d_sb, name); | |
1851 | if (!dentry) | |
1852 | return NULL; | |
a4464dbc AV |
1853 | spin_lock(&parent->d_lock); |
1854 | /* | |
1855 | * don't need child lock because it is not subject | |
1856 | * to concurrency here | |
1857 | */ | |
1858 | __dget_dlock(parent); | |
1859 | dentry->d_parent = parent; | |
946e51f2 | 1860 | list_add(&dentry->d_child, &parent->d_subdirs); |
a4464dbc AV |
1861 | spin_unlock(&parent->d_lock); |
1862 | ||
1863 | return dentry; | |
1864 | } | |
ec4f8605 | 1865 | EXPORT_SYMBOL(d_alloc); |
1da177e4 | 1866 | |
f9c34674 MS |
1867 | struct dentry *d_alloc_anon(struct super_block *sb) |
1868 | { | |
1869 | return __d_alloc(sb, NULL); | |
1870 | } | |
1871 | EXPORT_SYMBOL(d_alloc_anon); | |
1872 | ||
ba65dc5e AV |
1873 | struct dentry *d_alloc_cursor(struct dentry * parent) |
1874 | { | |
f9c34674 | 1875 | struct dentry *dentry = d_alloc_anon(parent->d_sb); |
ba65dc5e | 1876 | if (dentry) { |
5467a68c | 1877 | dentry->d_flags |= DCACHE_DENTRY_CURSOR; |
ba65dc5e AV |
1878 | dentry->d_parent = dget(parent); |
1879 | } | |
1880 | return dentry; | |
1881 | } | |
1882 | ||
e1a24bb0 BF |
1883 | /** |
1884 | * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems) | |
1885 | * @sb: the superblock | |
1886 | * @name: qstr of the name | |
1887 | * | |
1888 | * For a filesystem that just pins its dentries in memory and never | |
1889 | * performs lookups at all, return an unhashed IS_ROOT dentry. | |
5467a68c AV |
1890 | * This is used for pipes, sockets et.al. - the stuff that should |
1891 | * never be anyone's children or parents. Unlike all other | |
1892 | * dentries, these will not have RCU delay between dropping the | |
1893 | * last reference and freeing them. | |
ab1152dd AV |
1894 | * |
1895 | * The only user is alloc_file_pseudo() and that's what should | |
1896 | * be considered a public interface. Don't use directly. | |
e1a24bb0 | 1897 | */ |
4b936885 NP |
1898 | struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name) |
1899 | { | |
5467a68c AV |
1900 | struct dentry *dentry = __d_alloc(sb, name); |
1901 | if (likely(dentry)) | |
1902 | dentry->d_flags |= DCACHE_NORCU; | |
1903 | return dentry; | |
4b936885 | 1904 | } |
4b936885 | 1905 | |
1da177e4 LT |
1906 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) |
1907 | { | |
1908 | struct qstr q; | |
1909 | ||
1910 | q.name = name; | |
8387ff25 | 1911 | q.hash_len = hashlen_string(parent, name); |
1da177e4 LT |
1912 | return d_alloc(parent, &q); |
1913 | } | |
ef26ca97 | 1914 | EXPORT_SYMBOL(d_alloc_name); |
1da177e4 | 1915 | |
fb045adb NP |
1916 | void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) |
1917 | { | |
6f7f7caa LT |
1918 | WARN_ON_ONCE(dentry->d_op); |
1919 | WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH | | |
fb045adb NP |
1920 | DCACHE_OP_COMPARE | |
1921 | DCACHE_OP_REVALIDATE | | |
ecf3d1f1 | 1922 | DCACHE_OP_WEAK_REVALIDATE | |
4bacc9c9 | 1923 | DCACHE_OP_DELETE | |
d101a125 | 1924 | DCACHE_OP_REAL)); |
fb045adb NP |
1925 | dentry->d_op = op; |
1926 | if (!op) | |
1927 | return; | |
1928 | if (op->d_hash) | |
1929 | dentry->d_flags |= DCACHE_OP_HASH; | |
1930 | if (op->d_compare) | |
1931 | dentry->d_flags |= DCACHE_OP_COMPARE; | |
1932 | if (op->d_revalidate) | |
1933 | dentry->d_flags |= DCACHE_OP_REVALIDATE; | |
ecf3d1f1 JL |
1934 | if (op->d_weak_revalidate) |
1935 | dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE; | |
fb045adb NP |
1936 | if (op->d_delete) |
1937 | dentry->d_flags |= DCACHE_OP_DELETE; | |
f0023bc6 SW |
1938 | if (op->d_prune) |
1939 | dentry->d_flags |= DCACHE_OP_PRUNE; | |
d101a125 MS |
1940 | if (op->d_real) |
1941 | dentry->d_flags |= DCACHE_OP_REAL; | |
fb045adb NP |
1942 | |
1943 | } | |
1944 | EXPORT_SYMBOL(d_set_d_op); | |
1945 | ||
df1a085a DH |
1946 | |
1947 | /* | |
1948 | * d_set_fallthru - Mark a dentry as falling through to a lower layer | |
1949 | * @dentry - The dentry to mark | |
1950 | * | |
1951 | * Mark a dentry as falling through to the lower layer (as set with | |
1952 | * d_pin_lower()). This flag may be recorded on the medium. | |
1953 | */ | |
1954 | void d_set_fallthru(struct dentry *dentry) | |
1955 | { | |
1956 | spin_lock(&dentry->d_lock); | |
1957 | dentry->d_flags |= DCACHE_FALLTHRU; | |
1958 | spin_unlock(&dentry->d_lock); | |
1959 | } | |
1960 | EXPORT_SYMBOL(d_set_fallthru); | |
1961 | ||
b18825a7 DH |
1962 | static unsigned d_flags_for_inode(struct inode *inode) |
1963 | { | |
44bdb5e5 | 1964 | unsigned add_flags = DCACHE_REGULAR_TYPE; |
b18825a7 DH |
1965 | |
1966 | if (!inode) | |
1967 | return DCACHE_MISS_TYPE; | |
1968 | ||
1969 | if (S_ISDIR(inode->i_mode)) { | |
1970 | add_flags = DCACHE_DIRECTORY_TYPE; | |
1971 | if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) { | |
1972 | if (unlikely(!inode->i_op->lookup)) | |
1973 | add_flags = DCACHE_AUTODIR_TYPE; | |
1974 | else | |
1975 | inode->i_opflags |= IOP_LOOKUP; | |
1976 | } | |
44bdb5e5 DH |
1977 | goto type_determined; |
1978 | } | |
1979 | ||
1980 | if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) { | |
6b255391 | 1981 | if (unlikely(inode->i_op->get_link)) { |
b18825a7 | 1982 | add_flags = DCACHE_SYMLINK_TYPE; |
44bdb5e5 DH |
1983 | goto type_determined; |
1984 | } | |
1985 | inode->i_opflags |= IOP_NOFOLLOW; | |
b18825a7 DH |
1986 | } |
1987 | ||
44bdb5e5 DH |
1988 | if (unlikely(!S_ISREG(inode->i_mode))) |
1989 | add_flags = DCACHE_SPECIAL_TYPE; | |
1990 | ||
1991 | type_determined: | |
b18825a7 DH |
1992 | if (unlikely(IS_AUTOMOUNT(inode))) |
1993 | add_flags |= DCACHE_NEED_AUTOMOUNT; | |
1994 | return add_flags; | |
1995 | } | |
1996 | ||
360da900 OH |
1997 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) |
1998 | { | |
b18825a7 | 1999 | unsigned add_flags = d_flags_for_inode(inode); |
85c7f810 | 2000 | WARN_ON(d_in_lookup(dentry)); |
b18825a7 | 2001 | |
b23fb0a6 | 2002 | spin_lock(&dentry->d_lock); |
af0c9af1 WL |
2003 | /* |
2004 | * Decrement negative dentry count if it was in the LRU list. | |
2005 | */ | |
2006 | if (dentry->d_flags & DCACHE_LRU_LIST) | |
2007 | this_cpu_dec(nr_dentry_negative); | |
de689f5e | 2008 | hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry); |
a528aca7 | 2009 | raw_write_seqcount_begin(&dentry->d_seq); |
4bf46a27 | 2010 | __d_set_inode_and_type(dentry, inode, add_flags); |
a528aca7 | 2011 | raw_write_seqcount_end(&dentry->d_seq); |
affda484 | 2012 | fsnotify_update_flags(dentry); |
b23fb0a6 | 2013 | spin_unlock(&dentry->d_lock); |
360da900 OH |
2014 | } |
2015 | ||
1da177e4 LT |
2016 | /** |
2017 | * d_instantiate - fill in inode information for a dentry | |
2018 | * @entry: dentry to complete | |
2019 | * @inode: inode to attach to this dentry | |
2020 | * | |
2021 | * Fill in inode information in the entry. | |
2022 | * | |
2023 | * This turns negative dentries into productive full members | |
2024 | * of society. | |
2025 | * | |
2026 | * NOTE! This assumes that the inode count has been incremented | |
2027 | * (or otherwise set) by the caller to indicate that it is now | |
2028 | * in use by the dcache. | |
2029 | */ | |
2030 | ||
2031 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
2032 | { | |
946e51f2 | 2033 | BUG_ON(!hlist_unhashed(&entry->d_u.d_alias)); |
de689f5e | 2034 | if (inode) { |
b9680917 | 2035 | security_d_instantiate(entry, inode); |
873feea0 | 2036 | spin_lock(&inode->i_lock); |
de689f5e | 2037 | __d_instantiate(entry, inode); |
873feea0 | 2038 | spin_unlock(&inode->i_lock); |
de689f5e | 2039 | } |
1da177e4 | 2040 | } |
ec4f8605 | 2041 | EXPORT_SYMBOL(d_instantiate); |
1da177e4 | 2042 | |
1e2e547a AV |
2043 | /* |
2044 | * This should be equivalent to d_instantiate() + unlock_new_inode(), | |
2045 | * with lockdep-related part of unlock_new_inode() done before | |
2046 | * anything else. Use that instead of open-coding d_instantiate()/ | |
2047 | * unlock_new_inode() combinations. | |
2048 | */ | |
2049 | void d_instantiate_new(struct dentry *entry, struct inode *inode) | |
2050 | { | |
2051 | BUG_ON(!hlist_unhashed(&entry->d_u.d_alias)); | |
2052 | BUG_ON(!inode); | |
2053 | lockdep_annotate_inode_mutex_key(inode); | |
2054 | security_d_instantiate(entry, inode); | |
2055 | spin_lock(&inode->i_lock); | |
2056 | __d_instantiate(entry, inode); | |
2057 | WARN_ON(!(inode->i_state & I_NEW)); | |
c2b6d621 | 2058 | inode->i_state &= ~I_NEW & ~I_CREATING; |
1e2e547a AV |
2059 | smp_mb(); |
2060 | wake_up_bit(&inode->i_state, __I_NEW); | |
2061 | spin_unlock(&inode->i_lock); | |
2062 | } | |
2063 | EXPORT_SYMBOL(d_instantiate_new); | |
2064 | ||
adc0e91a AV |
2065 | struct dentry *d_make_root(struct inode *root_inode) |
2066 | { | |
2067 | struct dentry *res = NULL; | |
2068 | ||
2069 | if (root_inode) { | |
f9c34674 | 2070 | res = d_alloc_anon(root_inode->i_sb); |
5467a68c | 2071 | if (res) |
adc0e91a | 2072 | d_instantiate(res, root_inode); |
5467a68c | 2073 | else |
adc0e91a AV |
2074 | iput(root_inode); |
2075 | } | |
2076 | return res; | |
2077 | } | |
2078 | EXPORT_SYMBOL(d_make_root); | |
2079 | ||
f9c34674 MS |
2080 | static struct dentry *__d_instantiate_anon(struct dentry *dentry, |
2081 | struct inode *inode, | |
2082 | bool disconnected) | |
4ea3ada2 | 2083 | { |
9308a612 | 2084 | struct dentry *res; |
b18825a7 | 2085 | unsigned add_flags; |
4ea3ada2 | 2086 | |
f9c34674 | 2087 | security_d_instantiate(dentry, inode); |
873feea0 | 2088 | spin_lock(&inode->i_lock); |
d891eedb | 2089 | res = __d_find_any_alias(inode); |
9308a612 | 2090 | if (res) { |
873feea0 | 2091 | spin_unlock(&inode->i_lock); |
f9c34674 | 2092 | dput(dentry); |
9308a612 CH |
2093 | goto out_iput; |
2094 | } | |
2095 | ||
2096 | /* attach a disconnected dentry */ | |
1a0a397e BF |
2097 | add_flags = d_flags_for_inode(inode); |
2098 | ||
2099 | if (disconnected) | |
2100 | add_flags |= DCACHE_DISCONNECTED; | |
b18825a7 | 2101 | |
f9c34674 MS |
2102 | spin_lock(&dentry->d_lock); |
2103 | __d_set_inode_and_type(dentry, inode, add_flags); | |
2104 | hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry); | |
f1ee6162 | 2105 | if (!disconnected) { |
139351f1 LT |
2106 | hlist_bl_lock(&dentry->d_sb->s_roots); |
2107 | hlist_bl_add_head(&dentry->d_hash, &dentry->d_sb->s_roots); | |
2108 | hlist_bl_unlock(&dentry->d_sb->s_roots); | |
f1ee6162 | 2109 | } |
f9c34674 | 2110 | spin_unlock(&dentry->d_lock); |
873feea0 | 2111 | spin_unlock(&inode->i_lock); |
9308a612 | 2112 | |
f9c34674 | 2113 | return dentry; |
9308a612 CH |
2114 | |
2115 | out_iput: | |
2116 | iput(inode); | |
2117 | return res; | |
4ea3ada2 | 2118 | } |
1a0a397e | 2119 | |
f9c34674 MS |
2120 | struct dentry *d_instantiate_anon(struct dentry *dentry, struct inode *inode) |
2121 | { | |
2122 | return __d_instantiate_anon(dentry, inode, true); | |
2123 | } | |
2124 | EXPORT_SYMBOL(d_instantiate_anon); | |
2125 | ||
2126 | static struct dentry *__d_obtain_alias(struct inode *inode, bool disconnected) | |
2127 | { | |
2128 | struct dentry *tmp; | |
2129 | struct dentry *res; | |
2130 | ||
2131 | if (!inode) | |
2132 | return ERR_PTR(-ESTALE); | |
2133 | if (IS_ERR(inode)) | |
2134 | return ERR_CAST(inode); | |
2135 | ||
2136 | res = d_find_any_alias(inode); | |
2137 | if (res) | |
2138 | goto out_iput; | |
2139 | ||
2140 | tmp = d_alloc_anon(inode->i_sb); | |
2141 | if (!tmp) { | |
2142 | res = ERR_PTR(-ENOMEM); | |
2143 | goto out_iput; | |
2144 | } | |
2145 | ||
2146 | return __d_instantiate_anon(tmp, inode, disconnected); | |
2147 | ||
2148 | out_iput: | |
2149 | iput(inode); | |
2150 | return res; | |
2151 | } | |
2152 | ||
1a0a397e BF |
2153 | /** |
2154 | * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode | |
2155 | * @inode: inode to allocate the dentry for | |
2156 | * | |
2157 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
2158 | * similar open by handle operations. The returned dentry may be anonymous, | |
2159 | * or may have a full name (if the inode was already in the cache). | |
2160 | * | |
2161 | * When called on a directory inode, we must ensure that the inode only ever | |
2162 | * has one dentry. If a dentry is found, that is returned instead of | |
2163 | * allocating a new one. | |
2164 | * | |
2165 | * On successful return, the reference to the inode has been transferred | |
2166 | * to the dentry. In case of an error the reference on the inode is released. | |
2167 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
2168 | * be passed in and the error will be propagated to the return value, | |
2169 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
2170 | */ | |
2171 | struct dentry *d_obtain_alias(struct inode *inode) | |
2172 | { | |
f9c34674 | 2173 | return __d_obtain_alias(inode, true); |
1a0a397e | 2174 | } |
adc48720 | 2175 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 | 2176 | |
1a0a397e BF |
2177 | /** |
2178 | * d_obtain_root - find or allocate a dentry for a given inode | |
2179 | * @inode: inode to allocate the dentry for | |
2180 | * | |
2181 | * Obtain an IS_ROOT dentry for the root of a filesystem. | |
2182 | * | |
2183 | * We must ensure that directory inodes only ever have one dentry. If a | |
2184 | * dentry is found, that is returned instead of allocating a new one. | |
2185 | * | |
2186 | * On successful return, the reference to the inode has been transferred | |
2187 | * to the dentry. In case of an error the reference on the inode is | |
2188 | * released. A %NULL or IS_ERR inode may be passed in and will be the | |
2189 | * error will be propagate to the return value, with a %NULL @inode | |
2190 | * replaced by ERR_PTR(-ESTALE). | |
2191 | */ | |
2192 | struct dentry *d_obtain_root(struct inode *inode) | |
2193 | { | |
f9c34674 | 2194 | return __d_obtain_alias(inode, false); |
1a0a397e BF |
2195 | } |
2196 | EXPORT_SYMBOL(d_obtain_root); | |
2197 | ||
9403540c BN |
2198 | /** |
2199 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
2200 | * @inode: the inode case-insensitive lookup has found | |
2201 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
2202 | * @name: the case-exact name to be associated with the returned dentry | |
2203 | * | |
2204 | * This is to avoid filling the dcache with case-insensitive names to the | |
2205 | * same inode, only the actual correct case is stored in the dcache for | |
2206 | * case-insensitive filesystems. | |
2207 | * | |
3d742d4b RD |
2208 | * For a case-insensitive lookup match and if the case-exact dentry |
2209 | * already exists in the dcache, use it and return it. | |
9403540c BN |
2210 | * |
2211 | * If no entry exists with the exact case name, allocate new dentry with | |
2212 | * the exact case, and return the spliced entry. | |
2213 | */ | |
e45b590b | 2214 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
2215 | struct qstr *name) |
2216 | { | |
d9171b93 | 2217 | struct dentry *found, *res; |
9403540c | 2218 | |
b6520c81 CH |
2219 | /* |
2220 | * First check if a dentry matching the name already exists, | |
2221 | * if not go ahead and create it now. | |
2222 | */ | |
9403540c | 2223 | found = d_hash_and_lookup(dentry->d_parent, name); |
d9171b93 AV |
2224 | if (found) { |
2225 | iput(inode); | |
2226 | return found; | |
2227 | } | |
2228 | if (d_in_lookup(dentry)) { | |
2229 | found = d_alloc_parallel(dentry->d_parent, name, | |
2230 | dentry->d_wait); | |
2231 | if (IS_ERR(found) || !d_in_lookup(found)) { | |
2232 | iput(inode); | |
2233 | return found; | |
9403540c | 2234 | } |
d9171b93 AV |
2235 | } else { |
2236 | found = d_alloc(dentry->d_parent, name); | |
2237 | if (!found) { | |
2238 | iput(inode); | |
2239 | return ERR_PTR(-ENOMEM); | |
2240 | } | |
2241 | } | |
2242 | res = d_splice_alias(inode, found); | |
2243 | if (res) { | |
40a3cb0d | 2244 | d_lookup_done(found); |
d9171b93 AV |
2245 | dput(found); |
2246 | return res; | |
9403540c | 2247 | } |
4f522a24 | 2248 | return found; |
9403540c | 2249 | } |
ec4f8605 | 2250 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 | 2251 | |
4f48d5da XL |
2252 | /** |
2253 | * d_same_name - compare dentry name with case-exact name | |
2254 | * @parent: parent dentry | |
2255 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
2256 | * @name: the case-exact name to be associated with the returned dentry | |
2257 | * | |
2258 | * Return: true if names are same, or false | |
2259 | */ | |
2260 | bool d_same_name(const struct dentry *dentry, const struct dentry *parent, | |
2261 | const struct qstr *name) | |
12f8ad4b | 2262 | { |
d4c91a8f AV |
2263 | if (likely(!(parent->d_flags & DCACHE_OP_COMPARE))) { |
2264 | if (dentry->d_name.len != name->len) | |
2265 | return false; | |
2266 | return dentry_cmp(dentry, name->name, name->len) == 0; | |
12f8ad4b | 2267 | } |
6fa67e70 | 2268 | return parent->d_op->d_compare(dentry, |
d4c91a8f AV |
2269 | dentry->d_name.len, dentry->d_name.name, |
2270 | name) == 0; | |
12f8ad4b | 2271 | } |
4f48d5da | 2272 | EXPORT_SYMBOL_GPL(d_same_name); |
12f8ad4b | 2273 | |
ae2a8236 LT |
2274 | /* |
2275 | * This is __d_lookup_rcu() when the parent dentry has | |
2276 | * DCACHE_OP_COMPARE, which makes things much nastier. | |
2277 | */ | |
2278 | static noinline struct dentry *__d_lookup_rcu_op_compare( | |
2279 | const struct dentry *parent, | |
2280 | const struct qstr *name, | |
2281 | unsigned *seqp) | |
2282 | { | |
2283 | u64 hashlen = name->hash_len; | |
2284 | struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen)); | |
2285 | struct hlist_bl_node *node; | |
2286 | struct dentry *dentry; | |
2287 | ||
2288 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { | |
2289 | int tlen; | |
2290 | const char *tname; | |
2291 | unsigned seq; | |
2292 | ||
2293 | seqretry: | |
2294 | seq = raw_seqcount_begin(&dentry->d_seq); | |
2295 | if (dentry->d_parent != parent) | |
2296 | continue; | |
2297 | if (d_unhashed(dentry)) | |
2298 | continue; | |
2299 | if (dentry->d_name.hash != hashlen_hash(hashlen)) | |
2300 | continue; | |
2301 | tlen = dentry->d_name.len; | |
2302 | tname = dentry->d_name.name; | |
2303 | /* we want a consistent (name,len) pair */ | |
2304 | if (read_seqcount_retry(&dentry->d_seq, seq)) { | |
2305 | cpu_relax(); | |
2306 | goto seqretry; | |
2307 | } | |
2308 | if (parent->d_op->d_compare(dentry, tlen, tname, name) != 0) | |
2309 | continue; | |
2310 | *seqp = seq; | |
2311 | return dentry; | |
2312 | } | |
2313 | return NULL; | |
2314 | } | |
2315 | ||
31e6b01f NP |
2316 | /** |
2317 | * __d_lookup_rcu - search for a dentry (racy, store-free) | |
2318 | * @parent: parent dentry | |
2319 | * @name: qstr of name we wish to find | |
1f1e6e52 | 2320 | * @seqp: returns d_seq value at the point where the dentry was found |
31e6b01f NP |
2321 | * Returns: dentry, or NULL |
2322 | * | |
2323 | * __d_lookup_rcu is the dcache lookup function for rcu-walk name | |
2324 | * resolution (store-free path walking) design described in | |
2325 | * Documentation/filesystems/path-lookup.txt. | |
2326 | * | |
2327 | * This is not to be used outside core vfs. | |
2328 | * | |
2329 | * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock | |
2330 | * held, and rcu_read_lock held. The returned dentry must not be stored into | |
2331 | * without taking d_lock and checking d_seq sequence count against @seq | |
2332 | * returned here. | |
2333 | * | |
15570086 | 2334 | * A refcount may be taken on the found dentry with the d_rcu_to_refcount |
31e6b01f NP |
2335 | * function. |
2336 | * | |
2337 | * Alternatively, __d_lookup_rcu may be called again to look up the child of | |
2338 | * the returned dentry, so long as its parent's seqlock is checked after the | |
2339 | * child is looked up. Thus, an interlocking stepping of sequence lock checks | |
2340 | * is formed, giving integrity down the path walk. | |
12f8ad4b LT |
2341 | * |
2342 | * NOTE! The caller *has* to check the resulting dentry against the sequence | |
2343 | * number we've returned before using any of the resulting dentry state! | |
31e6b01f | 2344 | */ |
8966be90 LT |
2345 | struct dentry *__d_lookup_rcu(const struct dentry *parent, |
2346 | const struct qstr *name, | |
da53be12 | 2347 | unsigned *seqp) |
31e6b01f | 2348 | { |
26fe5750 | 2349 | u64 hashlen = name->hash_len; |
31e6b01f | 2350 | const unsigned char *str = name->name; |
8387ff25 | 2351 | struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen)); |
ceb5bdc2 | 2352 | struct hlist_bl_node *node; |
31e6b01f NP |
2353 | struct dentry *dentry; |
2354 | ||
2355 | /* | |
2356 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2357 | * required to prevent single threaded performance regressions | |
2358 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2359 | * Keep the two functions in sync. | |
2360 | */ | |
2361 | ||
ae2a8236 LT |
2362 | if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) |
2363 | return __d_lookup_rcu_op_compare(parent, name, seqp); | |
2364 | ||
31e6b01f NP |
2365 | /* |
2366 | * The hash list is protected using RCU. | |
2367 | * | |
2368 | * Carefully use d_seq when comparing a candidate dentry, to avoid | |
2369 | * races with d_move(). | |
2370 | * | |
2371 | * It is possible that concurrent renames can mess up our list | |
2372 | * walk here and result in missing our dentry, resulting in the | |
2373 | * false-negative result. d_lookup() protects against concurrent | |
2374 | * renames using rename_lock seqlock. | |
2375 | * | |
b0a4bb83 | 2376 | * See Documentation/filesystems/path-lookup.txt for more details. |
31e6b01f | 2377 | */ |
b07ad996 | 2378 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
8966be90 | 2379 | unsigned seq; |
31e6b01f | 2380 | |
12f8ad4b LT |
2381 | /* |
2382 | * The dentry sequence count protects us from concurrent | |
da53be12 | 2383 | * renames, and thus protects parent and name fields. |
12f8ad4b LT |
2384 | * |
2385 | * The caller must perform a seqcount check in order | |
da53be12 | 2386 | * to do anything useful with the returned dentry. |
12f8ad4b LT |
2387 | * |
2388 | * NOTE! We do a "raw" seqcount_begin here. That means that | |
2389 | * we don't wait for the sequence count to stabilize if it | |
2390 | * is in the middle of a sequence change. If we do the slow | |
2391 | * dentry compare, we will do seqretries until it is stable, | |
2392 | * and if we end up with a successful lookup, we actually | |
2393 | * want to exit RCU lookup anyway. | |
d4c91a8f AV |
2394 | * |
2395 | * Note that raw_seqcount_begin still *does* smp_rmb(), so | |
2396 | * we are still guaranteed NUL-termination of ->d_name.name. | |
12f8ad4b LT |
2397 | */ |
2398 | seq = raw_seqcount_begin(&dentry->d_seq); | |
31e6b01f NP |
2399 | if (dentry->d_parent != parent) |
2400 | continue; | |
2e321806 LT |
2401 | if (d_unhashed(dentry)) |
2402 | continue; | |
ae2a8236 LT |
2403 | if (dentry->d_name.hash_len != hashlen) |
2404 | continue; | |
2405 | if (dentry_cmp(dentry, str, hashlen_len(hashlen)) != 0) | |
2406 | continue; | |
da53be12 | 2407 | *seqp = seq; |
d4c91a8f | 2408 | return dentry; |
31e6b01f NP |
2409 | } |
2410 | return NULL; | |
2411 | } | |
2412 | ||
1da177e4 LT |
2413 | /** |
2414 | * d_lookup - search for a dentry | |
2415 | * @parent: parent dentry | |
2416 | * @name: qstr of name we wish to find | |
b04f784e | 2417 | * Returns: dentry, or NULL |
1da177e4 | 2418 | * |
b04f784e NP |
2419 | * d_lookup searches the children of the parent dentry for the name in |
2420 | * question. If the dentry is found its reference count is incremented and the | |
2421 | * dentry is returned. The caller must use dput to free the entry when it has | |
2422 | * finished using it. %NULL is returned if the dentry does not exist. | |
1da177e4 | 2423 | */ |
da2d8455 | 2424 | struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 | 2425 | { |
31e6b01f | 2426 | struct dentry *dentry; |
949854d0 | 2427 | unsigned seq; |
1da177e4 | 2428 | |
b8314f93 DY |
2429 | do { |
2430 | seq = read_seqbegin(&rename_lock); | |
2431 | dentry = __d_lookup(parent, name); | |
2432 | if (dentry) | |
1da177e4 LT |
2433 | break; |
2434 | } while (read_seqretry(&rename_lock, seq)); | |
2435 | return dentry; | |
2436 | } | |
ec4f8605 | 2437 | EXPORT_SYMBOL(d_lookup); |
1da177e4 | 2438 | |
31e6b01f | 2439 | /** |
b04f784e NP |
2440 | * __d_lookup - search for a dentry (racy) |
2441 | * @parent: parent dentry | |
2442 | * @name: qstr of name we wish to find | |
2443 | * Returns: dentry, or NULL | |
2444 | * | |
2445 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
2446 | * false-negative result due to unrelated rename activity. | |
2447 | * | |
2448 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
2449 | * however it must be used carefully, eg. with a following d_lookup in | |
2450 | * the case of failure. | |
2451 | * | |
2452 | * __d_lookup callers must be commented. | |
2453 | */ | |
a713ca2a | 2454 | struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 | 2455 | { |
1da177e4 | 2456 | unsigned int hash = name->hash; |
8387ff25 | 2457 | struct hlist_bl_head *b = d_hash(hash); |
ceb5bdc2 | 2458 | struct hlist_bl_node *node; |
31e6b01f | 2459 | struct dentry *found = NULL; |
665a7583 | 2460 | struct dentry *dentry; |
1da177e4 | 2461 | |
31e6b01f NP |
2462 | /* |
2463 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2464 | * required to prevent single threaded performance regressions | |
2465 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2466 | * Keep the two functions in sync. | |
2467 | */ | |
2468 | ||
b04f784e NP |
2469 | /* |
2470 | * The hash list is protected using RCU. | |
2471 | * | |
2472 | * Take d_lock when comparing a candidate dentry, to avoid races | |
2473 | * with d_move(). | |
2474 | * | |
2475 | * It is possible that concurrent renames can mess up our list | |
2476 | * walk here and result in missing our dentry, resulting in the | |
2477 | * false-negative result. d_lookup() protects against concurrent | |
2478 | * renames using rename_lock seqlock. | |
2479 | * | |
b0a4bb83 | 2480 | * See Documentation/filesystems/path-lookup.txt for more details. |
b04f784e | 2481 | */ |
1da177e4 LT |
2482 | rcu_read_lock(); |
2483 | ||
b07ad996 | 2484 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
1da177e4 | 2485 | |
1da177e4 LT |
2486 | if (dentry->d_name.hash != hash) |
2487 | continue; | |
1da177e4 LT |
2488 | |
2489 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
2490 | if (dentry->d_parent != parent) |
2491 | goto next; | |
d0185c08 LT |
2492 | if (d_unhashed(dentry)) |
2493 | goto next; | |
2494 | ||
d4c91a8f AV |
2495 | if (!d_same_name(dentry, parent, name)) |
2496 | goto next; | |
1da177e4 | 2497 | |
98474236 | 2498 | dentry->d_lockref.count++; |
d0185c08 | 2499 | found = dentry; |
1da177e4 LT |
2500 | spin_unlock(&dentry->d_lock); |
2501 | break; | |
2502 | next: | |
2503 | spin_unlock(&dentry->d_lock); | |
2504 | } | |
2505 | rcu_read_unlock(); | |
2506 | ||
2507 | return found; | |
2508 | } | |
2509 | ||
3e7e241f EB |
2510 | /** |
2511 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
2512 | * @dir: Directory to search in | |
2513 | * @name: qstr of name we wish to find | |
2514 | * | |
4f522a24 | 2515 | * On lookup failure NULL is returned; on bad name - ERR_PTR(-error) |
3e7e241f EB |
2516 | */ |
2517 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
2518 | { | |
3e7e241f EB |
2519 | /* |
2520 | * Check for a fs-specific hash function. Note that we must | |
2521 | * calculate the standard hash first, as the d_op->d_hash() | |
2522 | * routine may choose to leave the hash value unchanged. | |
2523 | */ | |
8387ff25 | 2524 | name->hash = full_name_hash(dir, name->name, name->len); |
fb045adb | 2525 | if (dir->d_flags & DCACHE_OP_HASH) { |
da53be12 | 2526 | int err = dir->d_op->d_hash(dir, name); |
4f522a24 AV |
2527 | if (unlikely(err < 0)) |
2528 | return ERR_PTR(err); | |
3e7e241f | 2529 | } |
4f522a24 | 2530 | return d_lookup(dir, name); |
3e7e241f | 2531 | } |
4f522a24 | 2532 | EXPORT_SYMBOL(d_hash_and_lookup); |
3e7e241f | 2533 | |
1da177e4 LT |
2534 | /* |
2535 | * When a file is deleted, we have two options: | |
2536 | * - turn this dentry into a negative dentry | |
2537 | * - unhash this dentry and free it. | |
2538 | * | |
2539 | * Usually, we want to just turn this into | |
2540 | * a negative dentry, but if anybody else is | |
2541 | * currently using the dentry or the inode | |
2542 | * we can't do that and we fall back on removing | |
2543 | * it from the hash queues and waiting for | |
2544 | * it to be deleted later when it has no users | |
2545 | */ | |
2546 | ||
2547 | /** | |
2548 | * d_delete - delete a dentry | |
2549 | * @dentry: The dentry to delete | |
2550 | * | |
2551 | * Turn the dentry into a negative dentry if possible, otherwise | |
2552 | * remove it from the hash queues so it can be deleted later | |
2553 | */ | |
2554 | ||
2555 | void d_delete(struct dentry * dentry) | |
2556 | { | |
c19457f0 | 2557 | struct inode *inode = dentry->d_inode; |
c19457f0 AV |
2558 | |
2559 | spin_lock(&inode->i_lock); | |
2560 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
2561 | /* |
2562 | * Are we the only user? | |
2563 | */ | |
98474236 | 2564 | if (dentry->d_lockref.count == 1) { |
13e3c5e5 | 2565 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
31e6b01f | 2566 | dentry_unlink_inode(dentry); |
c19457f0 | 2567 | } else { |
1da177e4 | 2568 | __d_drop(dentry); |
c19457f0 AV |
2569 | spin_unlock(&dentry->d_lock); |
2570 | spin_unlock(&inode->i_lock); | |
2571 | } | |
1da177e4 | 2572 | } |
ec4f8605 | 2573 | EXPORT_SYMBOL(d_delete); |
1da177e4 | 2574 | |
15d3c589 | 2575 | static void __d_rehash(struct dentry *entry) |
1da177e4 | 2576 | { |
15d3c589 | 2577 | struct hlist_bl_head *b = d_hash(entry->d_name.hash); |
61647823 | 2578 | |
1879fd6a | 2579 | hlist_bl_lock(b); |
b07ad996 | 2580 | hlist_bl_add_head_rcu(&entry->d_hash, b); |
1879fd6a | 2581 | hlist_bl_unlock(b); |
1da177e4 LT |
2582 | } |
2583 | ||
2584 | /** | |
2585 | * d_rehash - add an entry back to the hash | |
2586 | * @entry: dentry to add to the hash | |
2587 | * | |
2588 | * Adds a dentry to the hash according to its name. | |
2589 | */ | |
2590 | ||
2591 | void d_rehash(struct dentry * entry) | |
2592 | { | |
1da177e4 | 2593 | spin_lock(&entry->d_lock); |
15d3c589 | 2594 | __d_rehash(entry); |
1da177e4 | 2595 | spin_unlock(&entry->d_lock); |
1da177e4 | 2596 | } |
ec4f8605 | 2597 | EXPORT_SYMBOL(d_rehash); |
1da177e4 | 2598 | |
84e710da AV |
2599 | static inline unsigned start_dir_add(struct inode *dir) |
2600 | { | |
93f6d4e1 | 2601 | preempt_disable_nested(); |
84e710da AV |
2602 | for (;;) { |
2603 | unsigned n = dir->i_dir_seq; | |
2604 | if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n) | |
2605 | return n; | |
2606 | cpu_relax(); | |
2607 | } | |
2608 | } | |
2609 | ||
50417d22 SAS |
2610 | static inline void end_dir_add(struct inode *dir, unsigned int n, |
2611 | wait_queue_head_t *d_wait) | |
84e710da AV |
2612 | { |
2613 | smp_store_release(&dir->i_dir_seq, n + 2); | |
93f6d4e1 | 2614 | preempt_enable_nested(); |
50417d22 | 2615 | wake_up_all(d_wait); |
84e710da AV |
2616 | } |
2617 | ||
d9171b93 AV |
2618 | static void d_wait_lookup(struct dentry *dentry) |
2619 | { | |
2620 | if (d_in_lookup(dentry)) { | |
2621 | DECLARE_WAITQUEUE(wait, current); | |
2622 | add_wait_queue(dentry->d_wait, &wait); | |
2623 | do { | |
2624 | set_current_state(TASK_UNINTERRUPTIBLE); | |
2625 | spin_unlock(&dentry->d_lock); | |
2626 | schedule(); | |
2627 | spin_lock(&dentry->d_lock); | |
2628 | } while (d_in_lookup(dentry)); | |
2629 | } | |
2630 | } | |
2631 | ||
94bdd655 | 2632 | struct dentry *d_alloc_parallel(struct dentry *parent, |
d9171b93 AV |
2633 | const struct qstr *name, |
2634 | wait_queue_head_t *wq) | |
94bdd655 | 2635 | { |
94bdd655 | 2636 | unsigned int hash = name->hash; |
94bdd655 AV |
2637 | struct hlist_bl_head *b = in_lookup_hash(parent, hash); |
2638 | struct hlist_bl_node *node; | |
2639 | struct dentry *new = d_alloc(parent, name); | |
2640 | struct dentry *dentry; | |
2641 | unsigned seq, r_seq, d_seq; | |
2642 | ||
2643 | if (unlikely(!new)) | |
2644 | return ERR_PTR(-ENOMEM); | |
2645 | ||
2646 | retry: | |
2647 | rcu_read_lock(); | |
015555fd | 2648 | seq = smp_load_acquire(&parent->d_inode->i_dir_seq); |
94bdd655 AV |
2649 | r_seq = read_seqbegin(&rename_lock); |
2650 | dentry = __d_lookup_rcu(parent, name, &d_seq); | |
2651 | if (unlikely(dentry)) { | |
2652 | if (!lockref_get_not_dead(&dentry->d_lockref)) { | |
2653 | rcu_read_unlock(); | |
2654 | goto retry; | |
2655 | } | |
2656 | if (read_seqcount_retry(&dentry->d_seq, d_seq)) { | |
2657 | rcu_read_unlock(); | |
2658 | dput(dentry); | |
2659 | goto retry; | |
2660 | } | |
2661 | rcu_read_unlock(); | |
2662 | dput(new); | |
2663 | return dentry; | |
2664 | } | |
2665 | if (unlikely(read_seqretry(&rename_lock, r_seq))) { | |
2666 | rcu_read_unlock(); | |
2667 | goto retry; | |
2668 | } | |
015555fd WD |
2669 | |
2670 | if (unlikely(seq & 1)) { | |
2671 | rcu_read_unlock(); | |
2672 | goto retry; | |
2673 | } | |
2674 | ||
94bdd655 | 2675 | hlist_bl_lock(b); |
8cc07c80 | 2676 | if (unlikely(READ_ONCE(parent->d_inode->i_dir_seq) != seq)) { |
94bdd655 AV |
2677 | hlist_bl_unlock(b); |
2678 | rcu_read_unlock(); | |
2679 | goto retry; | |
2680 | } | |
94bdd655 AV |
2681 | /* |
2682 | * No changes for the parent since the beginning of d_lookup(). | |
2683 | * Since all removals from the chain happen with hlist_bl_lock(), | |
2684 | * any potential in-lookup matches are going to stay here until | |
2685 | * we unlock the chain. All fields are stable in everything | |
2686 | * we encounter. | |
2687 | */ | |
2688 | hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) { | |
2689 | if (dentry->d_name.hash != hash) | |
2690 | continue; | |
2691 | if (dentry->d_parent != parent) | |
2692 | continue; | |
d4c91a8f AV |
2693 | if (!d_same_name(dentry, parent, name)) |
2694 | continue; | |
94bdd655 | 2695 | hlist_bl_unlock(b); |
e7d6ef97 AV |
2696 | /* now we can try to grab a reference */ |
2697 | if (!lockref_get_not_dead(&dentry->d_lockref)) { | |
2698 | rcu_read_unlock(); | |
2699 | goto retry; | |
2700 | } | |
2701 | ||
2702 | rcu_read_unlock(); | |
2703 | /* | |
2704 | * somebody is likely to be still doing lookup for it; | |
2705 | * wait for them to finish | |
2706 | */ | |
d9171b93 AV |
2707 | spin_lock(&dentry->d_lock); |
2708 | d_wait_lookup(dentry); | |
2709 | /* | |
2710 | * it's not in-lookup anymore; in principle we should repeat | |
2711 | * everything from dcache lookup, but it's likely to be what | |
2712 | * d_lookup() would've found anyway. If it is, just return it; | |
2713 | * otherwise we really have to repeat the whole thing. | |
2714 | */ | |
2715 | if (unlikely(dentry->d_name.hash != hash)) | |
2716 | goto mismatch; | |
2717 | if (unlikely(dentry->d_parent != parent)) | |
2718 | goto mismatch; | |
2719 | if (unlikely(d_unhashed(dentry))) | |
2720 | goto mismatch; | |
d4c91a8f AV |
2721 | if (unlikely(!d_same_name(dentry, parent, name))) |
2722 | goto mismatch; | |
d9171b93 AV |
2723 | /* OK, it *is* a hashed match; return it */ |
2724 | spin_unlock(&dentry->d_lock); | |
94bdd655 AV |
2725 | dput(new); |
2726 | return dentry; | |
2727 | } | |
e7d6ef97 | 2728 | rcu_read_unlock(); |
94bdd655 AV |
2729 | /* we can't take ->d_lock here; it's OK, though. */ |
2730 | new->d_flags |= DCACHE_PAR_LOOKUP; | |
d9171b93 | 2731 | new->d_wait = wq; |
94bdd655 AV |
2732 | hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b); |
2733 | hlist_bl_unlock(b); | |
2734 | return new; | |
d9171b93 AV |
2735 | mismatch: |
2736 | spin_unlock(&dentry->d_lock); | |
2737 | dput(dentry); | |
2738 | goto retry; | |
94bdd655 AV |
2739 | } |
2740 | EXPORT_SYMBOL(d_alloc_parallel); | |
2741 | ||
45f78b0a SAS |
2742 | /* |
2743 | * - Unhash the dentry | |
2744 | * - Retrieve and clear the waitqueue head in dentry | |
2745 | * - Return the waitqueue head | |
2746 | */ | |
2747 | static wait_queue_head_t *__d_lookup_unhash(struct dentry *dentry) | |
85c7f810 | 2748 | { |
45f78b0a SAS |
2749 | wait_queue_head_t *d_wait; |
2750 | struct hlist_bl_head *b; | |
2751 | ||
2752 | lockdep_assert_held(&dentry->d_lock); | |
2753 | ||
2754 | b = in_lookup_hash(dentry->d_parent, dentry->d_name.hash); | |
94bdd655 | 2755 | hlist_bl_lock(b); |
85c7f810 | 2756 | dentry->d_flags &= ~DCACHE_PAR_LOOKUP; |
94bdd655 | 2757 | __hlist_bl_del(&dentry->d_u.d_in_lookup_hash); |
45f78b0a | 2758 | d_wait = dentry->d_wait; |
d9171b93 | 2759 | dentry->d_wait = NULL; |
94bdd655 AV |
2760 | hlist_bl_unlock(b); |
2761 | INIT_HLIST_NODE(&dentry->d_u.d_alias); | |
d9171b93 | 2762 | INIT_LIST_HEAD(&dentry->d_lru); |
45f78b0a SAS |
2763 | return d_wait; |
2764 | } | |
2765 | ||
2766 | void __d_lookup_unhash_wake(struct dentry *dentry) | |
2767 | { | |
2768 | spin_lock(&dentry->d_lock); | |
2769 | wake_up_all(__d_lookup_unhash(dentry)); | |
2770 | spin_unlock(&dentry->d_lock); | |
85c7f810 | 2771 | } |
45f78b0a | 2772 | EXPORT_SYMBOL(__d_lookup_unhash_wake); |
ed782b5a AV |
2773 | |
2774 | /* inode->i_lock held if inode is non-NULL */ | |
2775 | ||
2776 | static inline void __d_add(struct dentry *dentry, struct inode *inode) | |
2777 | { | |
45f78b0a | 2778 | wait_queue_head_t *d_wait; |
84e710da AV |
2779 | struct inode *dir = NULL; |
2780 | unsigned n; | |
0568d705 | 2781 | spin_lock(&dentry->d_lock); |
84e710da AV |
2782 | if (unlikely(d_in_lookup(dentry))) { |
2783 | dir = dentry->d_parent->d_inode; | |
2784 | n = start_dir_add(dir); | |
45f78b0a | 2785 | d_wait = __d_lookup_unhash(dentry); |
84e710da | 2786 | } |
ed782b5a | 2787 | if (inode) { |
0568d705 AV |
2788 | unsigned add_flags = d_flags_for_inode(inode); |
2789 | hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry); | |
2790 | raw_write_seqcount_begin(&dentry->d_seq); | |
2791 | __d_set_inode_and_type(dentry, inode, add_flags); | |
2792 | raw_write_seqcount_end(&dentry->d_seq); | |
affda484 | 2793 | fsnotify_update_flags(dentry); |
ed782b5a | 2794 | } |
15d3c589 | 2795 | __d_rehash(dentry); |
84e710da | 2796 | if (dir) |
50417d22 | 2797 | end_dir_add(dir, n, d_wait); |
0568d705 AV |
2798 | spin_unlock(&dentry->d_lock); |
2799 | if (inode) | |
2800 | spin_unlock(&inode->i_lock); | |
ed782b5a AV |
2801 | } |
2802 | ||
34d0d19d AV |
2803 | /** |
2804 | * d_add - add dentry to hash queues | |
2805 | * @entry: dentry to add | |
2806 | * @inode: The inode to attach to this dentry | |
2807 | * | |
2808 | * This adds the entry to the hash queues and initializes @inode. | |
2809 | * The entry was actually filled in earlier during d_alloc(). | |
2810 | */ | |
2811 | ||
2812 | void d_add(struct dentry *entry, struct inode *inode) | |
2813 | { | |
b9680917 AV |
2814 | if (inode) { |
2815 | security_d_instantiate(entry, inode); | |
ed782b5a | 2816 | spin_lock(&inode->i_lock); |
b9680917 | 2817 | } |
ed782b5a | 2818 | __d_add(entry, inode); |
34d0d19d AV |
2819 | } |
2820 | EXPORT_SYMBOL(d_add); | |
2821 | ||
668d0cd5 AV |
2822 | /** |
2823 | * d_exact_alias - find and hash an exact unhashed alias | |
2824 | * @entry: dentry to add | |
2825 | * @inode: The inode to go with this dentry | |
2826 | * | |
2827 | * If an unhashed dentry with the same name/parent and desired | |
2828 | * inode already exists, hash and return it. Otherwise, return | |
2829 | * NULL. | |
2830 | * | |
2831 | * Parent directory should be locked. | |
2832 | */ | |
2833 | struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode) | |
2834 | { | |
2835 | struct dentry *alias; | |
668d0cd5 AV |
2836 | unsigned int hash = entry->d_name.hash; |
2837 | ||
2838 | spin_lock(&inode->i_lock); | |
2839 | hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { | |
2840 | /* | |
2841 | * Don't need alias->d_lock here, because aliases with | |
2842 | * d_parent == entry->d_parent are not subject to name or | |
2843 | * parent changes, because the parent inode i_mutex is held. | |
2844 | */ | |
2845 | if (alias->d_name.hash != hash) | |
2846 | continue; | |
2847 | if (alias->d_parent != entry->d_parent) | |
2848 | continue; | |
d4c91a8f | 2849 | if (!d_same_name(alias, entry->d_parent, &entry->d_name)) |
668d0cd5 AV |
2850 | continue; |
2851 | spin_lock(&alias->d_lock); | |
2852 | if (!d_unhashed(alias)) { | |
2853 | spin_unlock(&alias->d_lock); | |
2854 | alias = NULL; | |
2855 | } else { | |
2856 | __dget_dlock(alias); | |
15d3c589 | 2857 | __d_rehash(alias); |
668d0cd5 AV |
2858 | spin_unlock(&alias->d_lock); |
2859 | } | |
2860 | spin_unlock(&inode->i_lock); | |
2861 | return alias; | |
2862 | } | |
2863 | spin_unlock(&inode->i_lock); | |
2864 | return NULL; | |
2865 | } | |
2866 | EXPORT_SYMBOL(d_exact_alias); | |
2867 | ||
8d85b484 | 2868 | static void swap_names(struct dentry *dentry, struct dentry *target) |
1da177e4 | 2869 | { |
8d85b484 AV |
2870 | if (unlikely(dname_external(target))) { |
2871 | if (unlikely(dname_external(dentry))) { | |
1da177e4 LT |
2872 | /* |
2873 | * Both external: swap the pointers | |
2874 | */ | |
9a8d5bb4 | 2875 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
2876 | } else { |
2877 | /* | |
2878 | * dentry:internal, target:external. Steal target's | |
2879 | * storage and make target internal. | |
2880 | */ | |
321bcf92 BF |
2881 | memcpy(target->d_iname, dentry->d_name.name, |
2882 | dentry->d_name.len + 1); | |
1da177e4 LT |
2883 | dentry->d_name.name = target->d_name.name; |
2884 | target->d_name.name = target->d_iname; | |
2885 | } | |
2886 | } else { | |
8d85b484 | 2887 | if (unlikely(dname_external(dentry))) { |
1da177e4 LT |
2888 | /* |
2889 | * dentry:external, target:internal. Give dentry's | |
2890 | * storage to target and make dentry internal | |
2891 | */ | |
2892 | memcpy(dentry->d_iname, target->d_name.name, | |
2893 | target->d_name.len + 1); | |
2894 | target->d_name.name = dentry->d_name.name; | |
2895 | dentry->d_name.name = dentry->d_iname; | |
2896 | } else { | |
2897 | /* | |
da1ce067 | 2898 | * Both are internal. |
1da177e4 | 2899 | */ |
da1ce067 MS |
2900 | unsigned int i; |
2901 | BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long))); | |
2902 | for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) { | |
2903 | swap(((long *) &dentry->d_iname)[i], | |
2904 | ((long *) &target->d_iname)[i]); | |
2905 | } | |
1da177e4 LT |
2906 | } |
2907 | } | |
a28ddb87 | 2908 | swap(dentry->d_name.hash_len, target->d_name.hash_len); |
1da177e4 LT |
2909 | } |
2910 | ||
8d85b484 AV |
2911 | static void copy_name(struct dentry *dentry, struct dentry *target) |
2912 | { | |
2913 | struct external_name *old_name = NULL; | |
2914 | if (unlikely(dname_external(dentry))) | |
2915 | old_name = external_name(dentry); | |
2916 | if (unlikely(dname_external(target))) { | |
2917 | atomic_inc(&external_name(target)->u.count); | |
2918 | dentry->d_name = target->d_name; | |
2919 | } else { | |
2920 | memcpy(dentry->d_iname, target->d_name.name, | |
2921 | target->d_name.len + 1); | |
2922 | dentry->d_name.name = dentry->d_iname; | |
2923 | dentry->d_name.hash_len = target->d_name.hash_len; | |
2924 | } | |
2925 | if (old_name && likely(atomic_dec_and_test(&old_name->u.count))) | |
2e03b4bc | 2926 | kfree_rcu(old_name, u.head); |
8d85b484 AV |
2927 | } |
2928 | ||
9eaef27b | 2929 | /* |
18367501 | 2930 | * __d_move - move a dentry |
1da177e4 LT |
2931 | * @dentry: entry to move |
2932 | * @target: new dentry | |
da1ce067 | 2933 | * @exchange: exchange the two dentries |
1da177e4 LT |
2934 | * |
2935 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2936 | * dcache entries should not be moved in this way. Caller must hold |
2937 | * rename_lock, the i_mutex of the source and target directories, | |
2938 | * and the sb->s_vfs_rename_mutex if they differ. See lock_rename(). | |
1da177e4 | 2939 | */ |
da1ce067 MS |
2940 | static void __d_move(struct dentry *dentry, struct dentry *target, |
2941 | bool exchange) | |
1da177e4 | 2942 | { |
42177007 | 2943 | struct dentry *old_parent, *p; |
45f78b0a | 2944 | wait_queue_head_t *d_wait; |
84e710da AV |
2945 | struct inode *dir = NULL; |
2946 | unsigned n; | |
1da177e4 | 2947 | |
42177007 AV |
2948 | WARN_ON(!dentry->d_inode); |
2949 | if (WARN_ON(dentry == target)) | |
2950 | return; | |
2951 | ||
2fd6b7f5 | 2952 | BUG_ON(d_ancestor(target, dentry)); |
42177007 AV |
2953 | old_parent = dentry->d_parent; |
2954 | p = d_ancestor(old_parent, target); | |
2955 | if (IS_ROOT(dentry)) { | |
2956 | BUG_ON(p); | |
2957 | spin_lock(&target->d_parent->d_lock); | |
2958 | } else if (!p) { | |
2959 | /* target is not a descendent of dentry->d_parent */ | |
2960 | spin_lock(&target->d_parent->d_lock); | |
2961 | spin_lock_nested(&old_parent->d_lock, DENTRY_D_LOCK_NESTED); | |
2962 | } else { | |
2963 | BUG_ON(p == dentry); | |
2964 | spin_lock(&old_parent->d_lock); | |
2965 | if (p != target) | |
2966 | spin_lock_nested(&target->d_parent->d_lock, | |
2967 | DENTRY_D_LOCK_NESTED); | |
2968 | } | |
2969 | spin_lock_nested(&dentry->d_lock, 2); | |
2970 | spin_lock_nested(&target->d_lock, 3); | |
2fd6b7f5 | 2971 | |
84e710da AV |
2972 | if (unlikely(d_in_lookup(target))) { |
2973 | dir = target->d_parent->d_inode; | |
2974 | n = start_dir_add(dir); | |
45f78b0a | 2975 | d_wait = __d_lookup_unhash(target); |
84e710da | 2976 | } |
1da177e4 | 2977 | |
31e6b01f | 2978 | write_seqcount_begin(&dentry->d_seq); |
1ca7d67c | 2979 | write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED); |
31e6b01f | 2980 | |
15d3c589 | 2981 | /* unhash both */ |
0632a9ac AV |
2982 | if (!d_unhashed(dentry)) |
2983 | ___d_drop(dentry); | |
2984 | if (!d_unhashed(target)) | |
2985 | ___d_drop(target); | |
1da177e4 | 2986 | |
076515fc AV |
2987 | /* ... and switch them in the tree */ |
2988 | dentry->d_parent = target->d_parent; | |
2989 | if (!exchange) { | |
8d85b484 | 2990 | copy_name(dentry, target); |
61647823 | 2991 | target->d_hash.pprev = NULL; |
076515fc | 2992 | dentry->d_parent->d_lockref.count++; |
5467a68c | 2993 | if (dentry != old_parent) /* wasn't IS_ROOT */ |
076515fc | 2994 | WARN_ON(!--old_parent->d_lockref.count); |
1da177e4 | 2995 | } else { |
076515fc AV |
2996 | target->d_parent = old_parent; |
2997 | swap_names(dentry, target); | |
946e51f2 | 2998 | list_move(&target->d_child, &target->d_parent->d_subdirs); |
076515fc AV |
2999 | __d_rehash(target); |
3000 | fsnotify_update_flags(target); | |
1da177e4 | 3001 | } |
076515fc AV |
3002 | list_move(&dentry->d_child, &dentry->d_parent->d_subdirs); |
3003 | __d_rehash(dentry); | |
3004 | fsnotify_update_flags(dentry); | |
0bf3d5c1 | 3005 | fscrypt_handle_d_move(dentry); |
1da177e4 | 3006 | |
31e6b01f NP |
3007 | write_seqcount_end(&target->d_seq); |
3008 | write_seqcount_end(&dentry->d_seq); | |
3009 | ||
84e710da | 3010 | if (dir) |
50417d22 | 3011 | end_dir_add(dir, n, d_wait); |
076515fc AV |
3012 | |
3013 | if (dentry->d_parent != old_parent) | |
3014 | spin_unlock(&dentry->d_parent->d_lock); | |
3015 | if (dentry != old_parent) | |
3016 | spin_unlock(&old_parent->d_lock); | |
3017 | spin_unlock(&target->d_lock); | |
3018 | spin_unlock(&dentry->d_lock); | |
18367501 AV |
3019 | } |
3020 | ||
3021 | /* | |
3022 | * d_move - move a dentry | |
3023 | * @dentry: entry to move | |
3024 | * @target: new dentry | |
3025 | * | |
3026 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
3027 | * dcache entries should not be moved in this way. See the locking |
3028 | * requirements for __d_move. | |
18367501 AV |
3029 | */ |
3030 | void d_move(struct dentry *dentry, struct dentry *target) | |
3031 | { | |
3032 | write_seqlock(&rename_lock); | |
da1ce067 | 3033 | __d_move(dentry, target, false); |
1da177e4 | 3034 | write_sequnlock(&rename_lock); |
9eaef27b | 3035 | } |
ec4f8605 | 3036 | EXPORT_SYMBOL(d_move); |
1da177e4 | 3037 | |
da1ce067 MS |
3038 | /* |
3039 | * d_exchange - exchange two dentries | |
3040 | * @dentry1: first dentry | |
3041 | * @dentry2: second dentry | |
3042 | */ | |
3043 | void d_exchange(struct dentry *dentry1, struct dentry *dentry2) | |
3044 | { | |
3045 | write_seqlock(&rename_lock); | |
3046 | ||
3047 | WARN_ON(!dentry1->d_inode); | |
3048 | WARN_ON(!dentry2->d_inode); | |
3049 | WARN_ON(IS_ROOT(dentry1)); | |
3050 | WARN_ON(IS_ROOT(dentry2)); | |
3051 | ||
3052 | __d_move(dentry1, dentry2, true); | |
3053 | ||
3054 | write_sequnlock(&rename_lock); | |
3055 | } | |
3056 | ||
e2761a11 OH |
3057 | /** |
3058 | * d_ancestor - search for an ancestor | |
3059 | * @p1: ancestor dentry | |
3060 | * @p2: child dentry | |
3061 | * | |
3062 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
3063 | * an ancestor of p2, else NULL. | |
9eaef27b | 3064 | */ |
e2761a11 | 3065 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
3066 | { |
3067 | struct dentry *p; | |
3068 | ||
871c0067 | 3069 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 3070 | if (p->d_parent == p1) |
e2761a11 | 3071 | return p; |
9eaef27b | 3072 | } |
e2761a11 | 3073 | return NULL; |
9eaef27b TM |
3074 | } |
3075 | ||
3076 | /* | |
3077 | * This helper attempts to cope with remotely renamed directories | |
3078 | * | |
3079 | * It assumes that the caller is already holding | |
a03e283b | 3080 | * dentry->d_parent->d_inode->i_mutex, and rename_lock |
9eaef27b TM |
3081 | * |
3082 | * Note: If ever the locking in lock_rename() changes, then please | |
3083 | * remember to update this too... | |
9eaef27b | 3084 | */ |
b5ae6b15 | 3085 | static int __d_unalias(struct inode *inode, |
873feea0 | 3086 | struct dentry *dentry, struct dentry *alias) |
9eaef27b | 3087 | { |
9902af79 AV |
3088 | struct mutex *m1 = NULL; |
3089 | struct rw_semaphore *m2 = NULL; | |
3d330dc1 | 3090 | int ret = -ESTALE; |
9eaef27b TM |
3091 | |
3092 | /* If alias and dentry share a parent, then no extra locks required */ | |
3093 | if (alias->d_parent == dentry->d_parent) | |
3094 | goto out_unalias; | |
3095 | ||
9eaef27b | 3096 | /* See lock_rename() */ |
9eaef27b TM |
3097 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) |
3098 | goto out_err; | |
3099 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
9902af79 | 3100 | if (!inode_trylock_shared(alias->d_parent->d_inode)) |
9eaef27b | 3101 | goto out_err; |
9902af79 | 3102 | m2 = &alias->d_parent->d_inode->i_rwsem; |
9eaef27b | 3103 | out_unalias: |
8ed936b5 | 3104 | __d_move(alias, dentry, false); |
b5ae6b15 | 3105 | ret = 0; |
9eaef27b | 3106 | out_err: |
9eaef27b | 3107 | if (m2) |
9902af79 | 3108 | up_read(m2); |
9eaef27b TM |
3109 | if (m1) |
3110 | mutex_unlock(m1); | |
3111 | return ret; | |
3112 | } | |
3113 | ||
3f70bd51 BF |
3114 | /** |
3115 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
3116 | * @inode: the inode which may have a disconnected dentry | |
3117 | * @dentry: a negative dentry which we want to point to the inode. | |
3118 | * | |
da093a9b BF |
3119 | * If inode is a directory and has an IS_ROOT alias, then d_move that in |
3120 | * place of the given dentry and return it, else simply d_add the inode | |
3121 | * to the dentry and return NULL. | |
3f70bd51 | 3122 | * |
908790fa BF |
3123 | * If a non-IS_ROOT directory is found, the filesystem is corrupt, and |
3124 | * we should error out: directories can't have multiple aliases. | |
3125 | * | |
3f70bd51 BF |
3126 | * This is needed in the lookup routine of any filesystem that is exportable |
3127 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
3128 | * | |
3129 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
3130 | * is returned. This matches the expected return value of ->lookup. | |
3131 | * | |
3132 | * Cluster filesystems may call this function with a negative, hashed dentry. | |
3133 | * In that case, we know that the inode will be a regular file, and also this | |
3134 | * will only occur during atomic_open. So we need to check for the dentry | |
3135 | * being already hashed only in the final case. | |
3136 | */ | |
3137 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
3138 | { | |
3f70bd51 BF |
3139 | if (IS_ERR(inode)) |
3140 | return ERR_CAST(inode); | |
3141 | ||
770bfad8 DH |
3142 | BUG_ON(!d_unhashed(dentry)); |
3143 | ||
de689f5e | 3144 | if (!inode) |
b5ae6b15 | 3145 | goto out; |
de689f5e | 3146 | |
b9680917 | 3147 | security_d_instantiate(dentry, inode); |
873feea0 | 3148 | spin_lock(&inode->i_lock); |
9eaef27b | 3149 | if (S_ISDIR(inode->i_mode)) { |
b5ae6b15 AV |
3150 | struct dentry *new = __d_find_any_alias(inode); |
3151 | if (unlikely(new)) { | |
a03e283b EB |
3152 | /* The reference to new ensures it remains an alias */ |
3153 | spin_unlock(&inode->i_lock); | |
18367501 | 3154 | write_seqlock(&rename_lock); |
b5ae6b15 AV |
3155 | if (unlikely(d_ancestor(new, dentry))) { |
3156 | write_sequnlock(&rename_lock); | |
b5ae6b15 AV |
3157 | dput(new); |
3158 | new = ERR_PTR(-ELOOP); | |
3159 | pr_warn_ratelimited( | |
3160 | "VFS: Lookup of '%s' in %s %s" | |
3161 | " would have caused loop\n", | |
3162 | dentry->d_name.name, | |
3163 | inode->i_sb->s_type->name, | |
3164 | inode->i_sb->s_id); | |
3165 | } else if (!IS_ROOT(new)) { | |
076515fc | 3166 | struct dentry *old_parent = dget(new->d_parent); |
b5ae6b15 | 3167 | int err = __d_unalias(inode, dentry, new); |
18367501 | 3168 | write_sequnlock(&rename_lock); |
b5ae6b15 AV |
3169 | if (err) { |
3170 | dput(new); | |
3171 | new = ERR_PTR(err); | |
3172 | } | |
076515fc | 3173 | dput(old_parent); |
18367501 | 3174 | } else { |
b5ae6b15 AV |
3175 | __d_move(new, dentry, false); |
3176 | write_sequnlock(&rename_lock); | |
dd179946 | 3177 | } |
b5ae6b15 AV |
3178 | iput(inode); |
3179 | return new; | |
9eaef27b | 3180 | } |
770bfad8 | 3181 | } |
b5ae6b15 | 3182 | out: |
ed782b5a | 3183 | __d_add(dentry, inode); |
b5ae6b15 | 3184 | return NULL; |
770bfad8 | 3185 | } |
b5ae6b15 | 3186 | EXPORT_SYMBOL(d_splice_alias); |
770bfad8 | 3187 | |
1da177e4 LT |
3188 | /* |
3189 | * Test whether new_dentry is a subdirectory of old_dentry. | |
3190 | * | |
3191 | * Trivially implemented using the dcache structure | |
3192 | */ | |
3193 | ||
3194 | /** | |
3195 | * is_subdir - is new dentry a subdirectory of old_dentry | |
3196 | * @new_dentry: new dentry | |
3197 | * @old_dentry: old dentry | |
3198 | * | |
a6e5787f YB |
3199 | * Returns true if new_dentry is a subdirectory of the parent (at any depth). |
3200 | * Returns false otherwise. | |
1da177e4 LT |
3201 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() |
3202 | */ | |
3203 | ||
a6e5787f | 3204 | bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 | 3205 | { |
a6e5787f | 3206 | bool result; |
949854d0 | 3207 | unsigned seq; |
1da177e4 | 3208 | |
e2761a11 | 3209 | if (new_dentry == old_dentry) |
a6e5787f | 3210 | return true; |
e2761a11 | 3211 | |
e2761a11 | 3212 | do { |
1da177e4 | 3213 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 3214 | seq = read_seqbegin(&rename_lock); |
949854d0 NP |
3215 | /* |
3216 | * Need rcu_readlock to protect against the d_parent trashing | |
3217 | * due to d_move | |
3218 | */ | |
3219 | rcu_read_lock(); | |
e2761a11 | 3220 | if (d_ancestor(old_dentry, new_dentry)) |
a6e5787f | 3221 | result = true; |
e2761a11 | 3222 | else |
a6e5787f | 3223 | result = false; |
949854d0 | 3224 | rcu_read_unlock(); |
1da177e4 | 3225 | } while (read_seqretry(&rename_lock, seq)); |
1da177e4 LT |
3226 | |
3227 | return result; | |
3228 | } | |
e8f9e5b7 | 3229 | EXPORT_SYMBOL(is_subdir); |
1da177e4 | 3230 | |
db14fc3a | 3231 | static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry) |
1da177e4 | 3232 | { |
db14fc3a MS |
3233 | struct dentry *root = data; |
3234 | if (dentry != root) { | |
3235 | if (d_unhashed(dentry) || !dentry->d_inode) | |
3236 | return D_WALK_SKIP; | |
1da177e4 | 3237 | |
01ddc4ed MS |
3238 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { |
3239 | dentry->d_flags |= DCACHE_GENOCIDE; | |
3240 | dentry->d_lockref.count--; | |
3241 | } | |
1da177e4 | 3242 | } |
db14fc3a MS |
3243 | return D_WALK_CONTINUE; |
3244 | } | |
58db63d0 | 3245 | |
db14fc3a MS |
3246 | void d_genocide(struct dentry *parent) |
3247 | { | |
3a8e3611 | 3248 | d_walk(parent, parent, d_genocide_kill); |
1da177e4 LT |
3249 | } |
3250 | ||
771eb4fe | 3251 | void d_mark_tmpfile(struct file *file, struct inode *inode) |
1da177e4 | 3252 | { |
863f144f MS |
3253 | struct dentry *dentry = file->f_path.dentry; |
3254 | ||
60545d0d | 3255 | BUG_ON(dentry->d_name.name != dentry->d_iname || |
946e51f2 | 3256 | !hlist_unhashed(&dentry->d_u.d_alias) || |
60545d0d AV |
3257 | !d_unlinked(dentry)); |
3258 | spin_lock(&dentry->d_parent->d_lock); | |
3259 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
3260 | dentry->d_name.len = sprintf(dentry->d_iname, "#%llu", | |
3261 | (unsigned long long)inode->i_ino); | |
3262 | spin_unlock(&dentry->d_lock); | |
3263 | spin_unlock(&dentry->d_parent->d_lock); | |
771eb4fe KO |
3264 | } |
3265 | EXPORT_SYMBOL(d_mark_tmpfile); | |
3266 | ||
3267 | void d_tmpfile(struct file *file, struct inode *inode) | |
3268 | { | |
3269 | struct dentry *dentry = file->f_path.dentry; | |
3270 | ||
3271 | inode_dec_link_count(inode); | |
3272 | d_mark_tmpfile(file, inode); | |
60545d0d | 3273 | d_instantiate(dentry, inode); |
1da177e4 | 3274 | } |
60545d0d | 3275 | EXPORT_SYMBOL(d_tmpfile); |
1da177e4 LT |
3276 | |
3277 | static __initdata unsigned long dhash_entries; | |
3278 | static int __init set_dhash_entries(char *str) | |
3279 | { | |
3280 | if (!str) | |
3281 | return 0; | |
3282 | dhash_entries = simple_strtoul(str, &str, 0); | |
3283 | return 1; | |
3284 | } | |
3285 | __setup("dhash_entries=", set_dhash_entries); | |
3286 | ||
3287 | static void __init dcache_init_early(void) | |
3288 | { | |
1da177e4 LT |
3289 | /* If hashes are distributed across NUMA nodes, defer |
3290 | * hash allocation until vmalloc space is available. | |
3291 | */ | |
3292 | if (hashdist) | |
3293 | return; | |
3294 | ||
3295 | dentry_hashtable = | |
3296 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3297 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3298 | dhash_entries, |
3299 | 13, | |
3d375d78 | 3300 | HASH_EARLY | HASH_ZERO, |
1da177e4 | 3301 | &d_hash_shift, |
b35d786b | 3302 | NULL, |
31fe62b9 | 3303 | 0, |
1da177e4 | 3304 | 0); |
854d3e63 | 3305 | d_hash_shift = 32 - d_hash_shift; |
1da177e4 LT |
3306 | } |
3307 | ||
74bf17cf | 3308 | static void __init dcache_init(void) |
1da177e4 | 3309 | { |
3d375d78 | 3310 | /* |
1da177e4 LT |
3311 | * A constructor could be added for stable state like the lists, |
3312 | * but it is probably not worth it because of the cache nature | |
3d375d78 | 3313 | * of the dcache. |
1da177e4 | 3314 | */ |
80344266 DW |
3315 | dentry_cache = KMEM_CACHE_USERCOPY(dentry, |
3316 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT, | |
3317 | d_iname); | |
1da177e4 LT |
3318 | |
3319 | /* Hash may have been set up in dcache_init_early */ | |
3320 | if (!hashdist) | |
3321 | return; | |
3322 | ||
3323 | dentry_hashtable = | |
3324 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3325 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3326 | dhash_entries, |
3327 | 13, | |
3d375d78 | 3328 | HASH_ZERO, |
1da177e4 | 3329 | &d_hash_shift, |
b35d786b | 3330 | NULL, |
31fe62b9 | 3331 | 0, |
1da177e4 | 3332 | 0); |
854d3e63 | 3333 | d_hash_shift = 32 - d_hash_shift; |
1da177e4 LT |
3334 | } |
3335 | ||
3336 | /* SLAB cache for __getname() consumers */ | |
68279f9c | 3337 | struct kmem_cache *names_cachep __ro_after_init; |
ec4f8605 | 3338 | EXPORT_SYMBOL(names_cachep); |
1da177e4 | 3339 | |
1da177e4 LT |
3340 | void __init vfs_caches_init_early(void) |
3341 | { | |
6916363f SAS |
3342 | int i; |
3343 | ||
3344 | for (i = 0; i < ARRAY_SIZE(in_lookup_hashtable); i++) | |
3345 | INIT_HLIST_BL_HEAD(&in_lookup_hashtable[i]); | |
3346 | ||
1da177e4 LT |
3347 | dcache_init_early(); |
3348 | inode_init_early(); | |
3349 | } | |
3350 | ||
4248b0da | 3351 | void __init vfs_caches_init(void) |
1da177e4 | 3352 | { |
6a9b8820 DW |
3353 | names_cachep = kmem_cache_create_usercopy("names_cache", PATH_MAX, 0, |
3354 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, 0, PATH_MAX, NULL); | |
1da177e4 | 3355 | |
74bf17cf DC |
3356 | dcache_init(); |
3357 | inode_init(); | |
4248b0da MG |
3358 | files_init(); |
3359 | files_maxfiles_init(); | |
74bf17cf | 3360 | mnt_init(); |
1da177e4 LT |
3361 | bdev_cache_init(); |
3362 | chrdev_init(); | |
3363 | } |